Abstract

Metacognition is a type of cognition: the awareness and understanding of one's own thought processes. It covers how people watch over and steer their own memory, decisions and learning, from judging whether a fact has been learned to sensing that an answer needs a second look. Researchers measure it by setting confidence against accuracy, which separates how closely confidence tracks performance from how high it runs overall. Accurate self-assessment matters for study, because learners act on their own judgments, and for mental health, where confidence and beliefs about one's own thinking shift in several disorders. Whether metacognition is one ability shared across memory, perception and reasoning, or a family of partly separate skills, remains an open question.

Keywords: metacognition, metamemory, confidence, judgments of learning, self-regulated learning

The word metacognition entered psychology through developmental research on how children keep track of their own memory and understanding; Flavell (1979) presented it as a new area of inquiry and argued that the ability to monitor one's own cognition grows over childhood. The field that followed is broad, spanning laboratory memory research, education, neuroscience and clinical psychology, and its central concepts are still not settled; an editorial opening a journal devoted to the topic listed a set of conceptual and methodological questions that remained unresolved (Veenman et al., 2006).

Everyday life is full of metacognitive moments. A person who cannot produce a colleague's name may still be certain that the name is known and about to surface. A student rereading notes the night before an examination decides, rightly or wrongly, that the material is now secure and stops studying. A driver unsure of a turn slows down and checks a sign. In each case the mind is judging its own state and acting on that judgment, and the quality of the outcome depends on whether the judgment is accurate.

Key Takeaways

  • This article divides metacognition into metacognitive knowledge, what a person knows about how thinking works; metacognitive monitoring, judging the state of one's own memory or performance; and metacognitive control, acting on those judgments.
  • Judgments of learning, predictions of how well something just studied will later be remembered, guide what people choose to study, but they follow cues such as font size that leave actual recall unchanged in experiments.
  • Calibration, whether average confidence matches average accuracy, and resolution, whether confidence is higher for right answers than wrong ones, are different properties and can come apart.
  • Meta-d′, a signal detection measure of how well confidence separates correct from incorrect decisions, is free of response bias, and its ratio to d′ gives metacognitive efficiency, sensitivity relative to task performance, although no current measure is wholly independent of task performance.
  • Whether metacognition is one general ability is contested: a meta-analysis of small studies found shared ability across perceptual tasks but no reliable link between memory and perception, and a study of 181 young adults with only 40 trials per task found it only partly general.
  • Metacognitive therapy, a treatment aimed at beliefs about one's own thinking, has shown large effects from before to after treatment and against waitlist in a meta-analysis of 25 studies, though the trials were few and small and allegiance bias is possible.

What Metacognition Is

Awareness and understanding of one's own thought processes.

Scope note for the MeSH descriptor Metacognition (D000067489), U.S. National Library of Medicine.

Metacognition is cognition directed at cognition. The term, also written meta-cognition, was used by Flavell (1979) as the label for a developmental research area concerned with cognitive monitoring, and his paper drew on earlier studies of preschool and elementary school children to argue that young children are poor at monitoring their own memory and comprehension; younger children in those studies tended to overestimate how ready they were to recall material they had studied.

This article uses three working parts. Metacognitive knowledge is what a person believes about minds and tasks, including their own strengths and the demands of a particular problem; Flavell (1979) treated it as a component of the field. Metacognitive monitoring (sometimes written meta-cognitive monitoring) is the ongoing assessment of one's own cognitive state, as when a learner judges whether a word pair has been learned or a decision-maker rates confidence in a choice. Metacognitive control (or meta-cognitive control) is the use of those assessments to regulate behaviour: deciding to restudy, to keep searching memory, to change an answer or to seek help. Monitoring without control would have no consequences, and control without monitoring would be blind.

Metacognition overlaps with, but is narrower than, awareness and consciousness in general: its object is specifically the person's own cognitive processes and products. From its early formulations, the concept has concerned how learners regulate their own cognitive activities while they learn (Veenman et al., 2006). In education, metacognitive awareness (also written meta-cognitive awareness) refers to a learner's reflective grasp of their own knowledge and regulation of learning, and researchers have developed ways of assessing it (Schraw & Dennison, 1994). Despite its wide use, the concept carries unresolved questions about definition and measurement, which an early editorial in the field set out as a list of open issues it did not claim was exhaustive (Veenman et al., 2006).

Monitoring and Control

The Nelson-Narens framework, set out by Nelson and Narens (1990) as a theoretical framework for metamemory, separates an object level that does the remembering or deciding from a meta level that receives information about it (monitoring) and sends instructions back to it (control) (Fleming & Dolan, 2012). The distinction remains supported as an organising scheme. Ackerman and Thompson (2017) carried the pairing of monitoring and control from memory into reasoning and problem solving, and Fleming's (2024) review of metacognition and confidence treats monitoring of one's own performance as a core part of the field.

The clearest test of whether monitoring actually drives control comes from study. A review by Metcalfe (2009) concluded that judgments of learning are not idle by-products of study but causally shape which items learners choose to restudy, drawing on experiments in which judgments of learning were shifted without any difference in learning and study choices shifted with them. The review also flagged that these were early results needing further scrutiny, and that in children the link between judgment and strategy was weaker.

How learners spend their time has been the main battleground. The discrepancy-reduction model held that people allocate more study time to items they judge difficult, and many of the experiments tallied by Son and Metcalfe (2000), as reported by Metcalfe (2009), did find more time going to harder items. That standing has since been revised. Son and Metcalfe (2000) noted that earlier studies had used short materials and generous time, and showed that when time was insufficient the preference reversed toward judged-easy items, while with shorter materials people favoured judged-difficult ones and under moderate time pressure showed no preference; Metcalfe (2009) likewise reported that the correlation between judged difficulty and study time disappeared when time was short.

The region of proximal learning was proposed by Metcalfe and Kornell (2005) as an account of study-time allocation. Its standing is that of a proposal supported in its original tests: Metcalfe's (2009) review placed it alongside the evidence that judgments of learning guide study choice and that learners do not always favour the hardest material, while cautioning that the effects of payoffs had not yet been explored and that individual differences were likely to be crucial. A third account, agenda-based regulation, holds that learners steer their study by a goal or agenda, such as the rewards on offer. Across four experiments that crossed reward with difficulty, people chose what to study according to the reward agenda more than according to how hard each item seemed (Ariel et al., 2009).

Metamemory

Metamemory, sometimes written meta-memory, is knowledge about and monitoring of one's own memory. An early demonstration came from Hart (1965), who in two small experiments (Ns of 22 and 16) asked people who had failed to recall an answer whether they nonetheless felt they knew it. Those feelings of knowing were fairly accurate indicators that the answer was stored, since they predicted later success on a recognition test.

Judgments of learning can be highly accurate under the right timing. Nelson and Dunlosky (1991) found that judgments made on the stimulus cue and delayed until shortly after study predicted subsequent recall extremely well, whereas judgments made immediately after study rarely did; they left the explanation of this delayed-JOL effect untested.

Why are such judgments sometimes accurate and sometimes not? The accessibility model of the feeling of knowing, proposed by Koriat (1993), holds that people do not read memory strength directly but infer knowing from how much partial information comes to mind, whether or not that information is correct. Its predictions were supported across three experiments, which counted against the view that people read an internal monitor directly (Koriat, 1993). A rival account held that the first sense of knowing comes from how familiar the words of the question seem, not from the answer itself (Reder & Ritter, 1992). In four experiments with paired associates, both feeling-of-knowing judgments and tip-of-the-tongue states tracked the familiarity of the cue, while the judgments bore no relation to how memorable the target was, which counted against the view that they come from retrieving the target itself; in one experiment a hypothesis based on everything retrieved, correct or not, fitted the results equally well (Metcalfe et al., 1993). A later set of experiments treated the two cues as partners rather than rivals: when Koriat and Levy-Sadot varied cue familiarity and accessibility independently of each other, using general information questions, each one lifted feeling-of-knowing judgments, and the authors joined them in a cascaded model in which accessibility counted chiefly once familiarity was already high (Koriat & Levy-Sadot, 2001). The cue-utilization approach of Koriat (1997) applied the same inferential logic to judgments of learning: across four paired-associate experiments, learners based their judgments on several kinds of cues and gave too little weight to extrinsic factors such as study conditions, and practice improved resolution while impairing calibration. The approach remains well supported, since later work shows judgments following cues that do not affect memory (Rhodes & Castel, 2008) and reviews describe learners misjudging their learning through a sense of fluency (Bjork et al., 2013).

One consequence is the illusion of competence. Koriat and Bjork (2005) showed in paired-associate experiments that judgments made while the answer is in view overrate how well it will be recalled later, when only the cue is available.

The font-size effect is the best-known illusion. Rhodes and Castel (2008) found that words in large type received higher judgments of learning than words in small type, although font size made no difference to free recall. Later work asked whether the font-size effect stems from the ease of reading large print or from people's beliefs about how memory works (Mueller et al., 2014). In the original study the effect persisted even when participants were warned, was modified when more valid cues were available, and disappeared when the large words were made less fluent to process, which Rhodes and Castel (2008) took as likely, though not proven, evidence that ease of processing drives it.

Demo 1: how well will you remember it?

You will see six word pairs, one at a time. For each, predict how likely you are to recall the second word when shown the first. Then take a short test.

Word pairs written for this article. Six items cannot show any one person's bias; the demo only shows the kind of comparison researchers make between predicted and actual recall.

Measuring Metacognition

Two properties of confidence must be kept apart. Calibration asks whether confidence is right on average: a person who says 80% across many answers should be correct about 80% of the time, and a gap in either direction is over- or underconfidence. Resolution, also called discrimination or sensitivity, asks whether confidence distinguishes one's own correct answers from one's errors, regardless of its overall level. A person can be poorly calibrated, with mean confidence above actual accuracy, yet still show good resolution by giving correct answers higher ratings than errors (Fleming & Lau, 2014).

The search for a good index of resolution has a long history. Nelson (1984) compared eight quantitative measures of the accuracy of feeling-of-knowing predictions and concluded that the Goodman-Kruskal gamma correlation seemed the best of them, while the Hart difference score was flawed. Later methodological work showed that gamma and the phi correlation are themselves contaminated by response bias, so a person who simply uses high ratings more often can appear more or less metacognitive than they are (Fleming & Lau, 2014).

Signal detection theory offered a way out. Maniscalco and Lau (2012) introduced meta-d′, which expresses how well confidence ratings separate correct from incorrect decisions in the same units as d′, the sensitivity of the decision itself; tested on a two-interval forced-choice visual task, participants came close to, but fell below, the level that their first-order sensitivity would allow, and the authors argued that older second-order measures were influenced by response bias and type 1 sensitivity. As a measurement framework, signal detection theory is now well supported: a methodological review recommended bias-free measures over simple correlations and preferred meta-d′, while warning that low metacognitive sensitivity does not by itself prove a lack of awareness (Fleming & Lau, 2014). A later secondary analysis of the Confidence Database (N = 6912) by Guggenmos (2021) found that every current measure of metacognitive performance kept some dependence on type 1 performance. Test-retest reliability varied with the number of trials, and typical studies often ran too few trials (Guggenmos, 2021).

Because meta-d′ is in the units of d′, the two can be compared directly. Metacognitive efficiency, the ratio meta-d′/d′, asks how much of the information available to the decision also reaches confidence; a ratio of 1 means confidence uses all of it (Fleming & Lau, 2014). The ratio does not reveal why efficiency falls short, whether through noise added to confidence or through other causes.

Demo 2: separating confidence accuracy from task accuracy

low confidencehigh confidencegreen: correct trials · red: wrong trials
Accuracy 76% · d′ 1.5 · estimated meta-d′ 0.7 · ratio 0.47
Extra noise blurs confidence, so the two histograms overlap more: meta-d′ falls below d′ even though accuracy has not changed.

Illustrative simulation of 3000 trials under an equal-variance signal detection model. Meta-d′ is estimated here by matching the area under the type 2 curve to a noise-free observer, a shortcut; published work fits the full model.

Measurement has also become a shared enterprise. The Confidence Database brings together 145 datasets, close to 4 million trials from over 8,700 people, and stays open to new contributions (Rahnev et al., 2020).

Better measures sharpened an old question: is metacognition domain-general, one ability that serves every task, or a set of task-specific skills? The issue is contested. A review with meta-analysis covering 12 manuscripts and 19 experiments found that metacognitive efficiency was shared across perceptual tasks but found no significant relation between memory and perception, with the caveat that the studies were small and low in power and used mixed metrics (Rouault et al., 2018b). A study of 181 young adults, mostly women at one university, compared four tasks and, in analyses that departed from its preregistered plan, concluded that metacognitive efficiency is partly domain-general, but each task had only 40 trials, one pairing's estimate included zero, and domain-specific processes were also involved (Mazancieux et al., 2020).

Confidence and Decisions

Confidence links metacognition to the wider study of decision making. Fleming's (2024) review argued that propositional confidence in one's own decisions or actions connects the neuroscience of uncertainty with the psychology of metacognition, and that such confidence draws on models the observer holds of the world and of their own mind, which is why judgments about one's own performance are inferential and can part company with it.

Demo 3: how well does your confidence track your accuracy?

Two boxes of dots flash for under a second. Pick the box with more dots, then say how confident you are. There are 12 trials.

A short task written for this article in the style of perceptual confidence experiments. Twelve trials give a rough picture only; research studies use hundreds of trials per person.

Confidence signals are not unique to humans. Kepecs and colleagues recorded single neurons in the orbitofrontal cortex of rats performing an odour categorization task and found activity that tracked decision confidence; combining behaviour, recordings and computational analysis, they showed that rats' willingness to wait for a delayed reward increased with confidence (Kepecs et al., 2008).

In humans, poor performance can bring poor self-assessment. The Dunning-Kruger effect comes from four studies using tests of humour, grammar and logic, in which participants in the bottom quartile, scoring around the 12th percentile, estimated their standing near the 62nd; the authors attributed this miscalibration to a metacognitive deficit, although that deficit explained the overestimation only in part (Kruger & Dunning, 1999). The effect is now contested: Gignac and Zajenkowski (2020) titled their paper as an argument that it is mostly a statistical artefact and addressed how the hypothesis should be tested with individual-differences data.

Miscalibration may also carry costs for learning. Dunlosky and Rawson (2012) is titled as arguing that overconfidence leads students to underachieve because inaccurate self-evaluation harms their learning and retention.

Metacognition in the Brain

People differ in how well their confidence tracks their accuracy, and those differences have a structural correlate. Fleming and colleagues measured introspective accuracy on a contrast detection task and related it to brain structure, reporting that more accurate individuals had more gray matter in the anterior prefrontal cortex and differences in white matter of the corpus callosum (Fleming et al., 2010).

A review by Fleming and Dolan (2012) linked lateral prefrontal regions, rostrolateral and dorsolateral, to the accuracy of retrospective confidence judgments, and medial regions to prospective judgments such as feelings of knowing. The authors noted that most of the evidence was correlational, that neural studies often confounded metacognition with task performance, and that whether the ability is domain-general was unanswered.

Pooling imaging studies gives a broader view. A coordinate-based meta-analysis of 47 studies with 2215 healthy participants found a network for confidence judgments, including prefrontal cortex, precuneus and insula, that appeared across domains; results for metacognitive sensitivity specifically were inconclusive because few studies addressed it (Vaccaro & Fleming, 2018).

Development and Other Species

Metacognitive skill changes across the lifespan. The early developmental work summarised by Flavell (1979) showed that young children monitor their memory poorly. A later cross-sectional study of 56 participants aged 11 to 41, using a perceptual task, found that metacognitive ability rose through adolescence and then levelled off in adulthood; the sample was small, only one task was used, and individuals were not followed over time (Weil et al., 2013).

Whether non-human animals have metacognition is debated. A review of uncertainty monitoring across humans, monkeys and one dolphin concluded that the animals' use of an uncertain response on perceptual and memory tasks resembled human performance, while leaving open whether this implies conscious self-awareness (Smith et al., 2003). Rhesus macaques previously trained to make retrospective confidence judgments on perceptual tasks carried them over at once both to an unfamiliar perceptual task and to a test of working memory, and also learned to request hints (Kornell et al., 2007).

The skeptical reading holds that such data can be explained without any representation of the animal's own mental states. Carruthers (2008) argued that first-order processes, graded beliefs and desires about the world, account for the animal findings, an explanation that assumes beliefs and desires come in different strengths, or degrees. The disagreement turns on what counts as evidence for monitoring one's own mind rather than responding to the world.

Metacognition and Learning

Learners often hold mistaken beliefs about how learning works. A review by Bjork and colleagues argued that people often misjudge how they learn and remember, misassessing and mismanaging their own learning, with their judgments swayed by a subjective sense of fluency (Bjork et al., 2013).

Category learning shows the gap vividly. In experiments asking people to learn painters' styles from examples, spacing the paintings of each artist among others produced better learning than presenting them in blocks, contradicting the authors' own prediction, yet most learners judged the blocked schedule more effective (Kornell & Bjork, 2008).

Such findings sit within the study of self-regulated learning, the planning, monitoring and adjustment of one's own learning, which Zimmerman (2002) surveyed in an overview of becoming a self-regulated learner. The MASRL model of Efklides (2011) adds motivation and affect, distinguishing a stable person level from a task-specific level at which metacognitive experiences such as a feeling of difficulty operate, with regulation running both top-down and bottom-up. Efklides (2011) stated its relations as hypotheses rather than demonstrated findings.

Training in self-regulation appears to help. Two meta-analyses, of 49 primary-school and 35 secondary-school studies, found that interventions improved academic performance, strategy use and motivation, with larger effects when researchers rather than teachers delivered the training and in mathematics compared with other subjects (Dignath & Büttner, 2008). An earlier meta-analysis of 51 studies of learning skills interventions concluded that, except for simple memorisation, training should be set in the context of the subject being taught (Hattie et al., 1996). Applied to learning in schools, these results suggest that metacognitive skills are teachable, though effects vary with how and by whom they are taught.

Metacognition and Reasoning

Reasoning has its own metacognitive signal. In four experiments with conditional, syllogistic and base-rate problems, Thompson and colleagues asked people for a quick first answer and a rating of how right it felt; a low feeling of rightness predicted longer rethinking and a greater chance of changing the answer, and answers that came to mind fluently felt more right (Thompson et al., 2011). The feeling thus decides when intuition is accepted and when slower analysis takes over. That time course has been challenged: in four experiments using a two-response paradigm with a response deadline and a concurrent load, people frequently gave the correct logical answer already as their first, fast response (Bago & De Neys, 2017).

The meta-reasoning framework of Ackerman and Thompson (2017) generalises this, proposing that monitoring judgments such as feelings of certainty regulate how much effort people invest in thinking and problem solving. The framework is supported by work on the feeling of rightness, though its authors caution that the heuristic cues behind these judgments are not necessarily reliable and that little research has looked beyond memory tasks.

Metacognition and Mental Health

Beliefs about one's own thinking can sustain psychological disorder. The metacognitive model of generalized anxiety disorder, developed by Wells (1995), holds that people with the condition hold positive beliefs about worry as a way of coping, come to appraise worry itself as dangerous (worry about worry), and make counterproductive attempts to control it. The paper illustrated treatment with a single case. Treatments that target the beliefs it names have been tested in trials: a meta-analysis of metacognitive therapy covering 25 studies and 780 adult patients found it effective for anxiety and depression and tentatively suggested it may outperform cognitive behavioural therapy, but the trials were few and small, risk of bias was often unclear or high, allegiance bias was possible, and the advantage over cognitive behavioural therapy was smaller at follow-up, where the lower bound of its confidence interval was close to zero (Normann & Morina, 2018).

Such beliefs can be measured. The MCQ-30, a 30-item short form of the Metacognitions Questionnaire, assesses several kinds of belief about one's own thinking organised into five factors, with good psychometric properties overall, although test-retest reliability ranged only from acceptable to good and model fit was acceptable rather than excellent (Wells & Cartwright-Hatton, 2004).

In schizophrenia, metacognitive training targets thinking biases such as leaping to conclusions and excessive certainty in errors. A narrative review reported preliminary support for it and suggested combining it with individual cognitive behavioural therapy (Moritz & Woodward, 2007). A later meta-analysis of 12 randomised trials with 807 participants with schizophrenia spectrum disorders found medium gains in cognitive insight, measured by self-report, but clinical insight could not be meta-analysed, heterogeneity was high on one index, follow-up lasted only 6 months, and the gain in self-certainty had gone by follow-up (Lopez-Morinigo et al., 2020).

Confidence itself is altered across conditions. A systematic review of 83 studies found confidence abnormalities in obsessive-compulsive disorder, schizophrenia, addiction, anxiety and depression, suggesting a transdiagnostic pattern, although the evidence for addiction, depression and anxiety was only suggestive, some obsessive-compulsive studies found no difference, and confidence was often confounded with task performance (Hoven et al., 2019). In an online study of 995 adults completing a perceptual task, self-reported symptom dimensions, in a general-population sample with no diagnostic information recorded, were associated with dissociable shifts in confidence and metacognition while task performance was unaffected; effects on metacognitive efficiency were uncorrected and need cautious reading (Rouault et al., 2018a). Building on this line, Seow and colleagues proposed separating local metacognition, confidence in single decisions, from global metacognition, broader beliefs about one's abilities, arguing that the global level has been neglected in psychiatric research (Seow et al., 2021).

Metacognition in Social Life

Metacognition also serves communication. Frith (2012) argued that explicit metacognition, the kind that can be put into words, is fundamentally social, letting people share and combine their confidence to make better joint decisions; he noted that such pairs gain an advantage only when their abilities are similar, that introspective access is very limited, and that his claim that explicit metacognition is uniquely human was his own belief. This connects metacognition to theory of mind, the capacity to think about other people's mental states.

Supra-personal cognitive control extends the idea. Shea and colleagues proposed that an explicit form of metacognition lets groups regulate one another's cognition, supporting cooperation and cumulative culture (Shea et al., 2014). The proposal remains speculative: its authors described the evidence for it as limited and preliminary.

Meta-emotion, also written metaemotion, applies the same reflexive logic to feeling. Gottman and colleagues described parents' meta-emotion philosophy, their feelings and beliefs about emotion in themselves and their children, and in a longitudinal study following children from age 5 to age 8 linked it through path analysis to parenting and to children's later outcomes and regulatory physiology; the data were described as preliminary (Gottman et al., 1996).

Worked Example

The following example is illustrative: the numbers are invented for this article and come from no study. One imaginary student answers eight quiz questions and, after each, rates confidence between 50% (a guess) and 100% (certain). In order, the confidences and outcomes are 90 correct, 80 correct, 80 wrong, 70 correct, 60 correct, 60 wrong, 50 wrong and 90 correct.

Calibration compares overall confidence with overall accuracy. The eight confidences sum to 580, so mean confidence is 72.5%. Five of the eight answers are correct, an accuracy of 62.5%. Mean confidence exceeds accuracy by 10 points, so the student is overconfident.

Resolution asks a different question: is confidence higher when the student is right? The five correct answers carry confidences of 90, 80, 70, 60 and 90, a mean of 78%. The three wrong answers carry 80, 60 and 50, a mean of about 63%. Confidence is about 15 points higher for correct answers, so resolution is good: the student's confidence carries real information about which answers to trust.

The two measures diverge. A teacher looking only at calibration would conclude that the student's self-assessment is poor; one looking only at resolution would conclude that it is good. Both are right about different things. The student knows which answers are shaky but sets the overall level too high, a pattern that could be corrected by lowering every rating without changing their order, whereas a student with poor resolution would need to learn which answers to doubt in the first place.

Table 1

Illustrative Confidence Ratings for One Imaginary Student's Eight Quiz Answers

QuestionConfidence (%)Answer
190Correct
280Correct
380Wrong
470Correct
560Correct
660Wrong
750Wrong
890Correct
Mean confidence72.5Accuracy 62.5% (5 of 8)
Mean confidence, correct answers78Calibration gap: +10 points
Mean confidence, wrong answersabout 63Resolution: higher when right

Note. This article's own invented numbers, not data. Calibration compares average confidence with accuracy; resolution compares confidence on correct and wrong answers.

Discussion

Figure 1

Monitoring and Control Between a Meta Level and an Object Level

A meta level above an object level, joined by monitoring flowing up and control flowing down A diagram with two boxes. The upper box, the meta level, holds a dashed inner box labelled model of the task. The lower box, the object level, is labelled remembering, deciding and learning. A green arrow on the left points up from the object level to the meta level and is labelled monitoring, with examples: judgments of learning, feelings of knowing, confidence. An amber arrow on the right points down from the meta level to the object level and is labelled control, with examples: restudy an item, stop searching memory, change an answer. meta level model of the task object level remembering, deciding, learning monitoring judgments of learning, feelings of knowing, confidence control restudy an item, stop searching memory, change an answer
Note. An original schematic of the monitoring-and-control idea, not data.

Several conclusions about metacognition are reasonably well supported. Confidence and judgments of learning track performance, but imperfectly, and they are built from cues: some, such as the delay since study, make them more accurate (Nelson & Dunlosky, 1991), while others, such as font size, inflate them without any benefit to memory (Rhodes & Castel, 2008). Monitoring appears to guide control, since judgments of learning are causally related to what people choose to study, although the review called these early results that invite further scrutiny (Metcalfe, 2009). And confidence can now be measured apart from response bias, and metacognitive efficiency, sensitivity measured relative to task performance, can be compared across domains and groups whose performance differs (Fleming & Lau, 2014), although no current measure is fully independent of task performance (Guggenmos, 2021).

Other questions remain open. Whether metacognition is one ability or many is unresolved: shared variance appears within perception, while evidence for a link between memory and perceptual metacognition is mixed, absent in a meta-analysis of small studies (Rouault et al., 2018b) but present for semantic, though not episodic, memory in a larger study (Mazancieux et al., 2020). How the brain computes confidence is still being worked out; imaging points to a common prefrontal, precuneus and insula network for confidence judgments, but findings for sensitivity itself are inconclusive (Vaccaro & Fleming, 2018). How far the animal findings require true metacognition rather than first-order processes is disputed (Carruthers, 2008). And the clinical evidence, though promising, rests on few and small trials, so which effects will hold up in larger, independent studies is not yet known (Normann & Morina, 2018).

The designs that would settle these questions are clear in outline: many trials per participant across several tasks with matched formats, longitudinal rather than cross-sectional developmental samples, imaging studies that hold task performance constant, and large clinical trials run by groups without a stake in the treatment.

Current Directions

Recent work has moved toward open data and stronger measurement. The Confidence Database makes thousands of participants' trial-level confidence ratings freely available for reanalysis (Rahnev et al., 2020), and hierarchical Bayesian estimation of metacognitive efficiency has been used to test domain generality across tasks (Mazancieux et al., 2020), extending earlier meta-analytic work on whether efficiency is shared across domains (Rouault et al., 2018b).

Computational psychiatry has taken up confidence as a target. Large online samples relate symptom dimensions to metacognition rather than to performance (Rouault et al., 2018a), systematic reviews map confidence abnormalities across diagnoses (Hoven et al., 2019), and the distinction between local confidence and global self-beliefs offers a way to connect laboratory measures with clinical experience (Seow et al., 2021).

Synthesis is also advancing. A recent review positions confidence as the bridge between decision neuroscience and metacognition (Fleming, 2024), and meta-analysis of imaging studies has identified a network common to confidence judgments across domains (Vaccaro & Fleming, 2018). On the applied side, meta-analyses have assessed metacognitive therapy (Normann & Morina, 2018) and metacognitive interventions for insight in schizophrenia spectrum disorders (Lopez-Morinigo et al., 2020). Research on reasoning has extended monitoring and control from memory to problem solving (Ackerman & Thompson, 2017), and the Dunning-Kruger effect remains under dispute as a possible statistical artefact (Gignac & Zajenkowski, 2020).

Common Misconceptions

Feeling sure that something will be remembered means it will be.
Judgments of learning respond to cues that have nothing to do with memory. Words printed in large type are judged more memorable than words in small type, yet they are recalled no better, and the illusion survives a warning (Rhodes & Castel, 2008).
A study method that feels effective is effective.
When people learned painters' styles, interleaving the paintings produced better learning than blocking them, but most learners rated blocking as the better method (Kornell & Bjork, 2008). Blocked study apparently creates a sense of fluent learning, and that sense can be mistaken for learning itself.
Someone who is good at judging their own memory is equally good at judging their own perception.
A meta-analysis found metacognitive efficiency shared across perceptual tasks but no significant relation between memory and perceptual metacognition, although the pooled studies were small (Rouault et al., 2018b). The assumption of a single general ability is natural, but the evidence for it is mixed.
Only humans can monitor their own uncertainty.
Monkeys and a dolphin use an uncertain response on difficult trials in a way that resembles human uncertainty monitoring (Smith et al., 2003). Whether this reflects metacognition or simpler first-order processes is debated (Carruthers, 2008), but the behaviour itself is not uniquely human.

Glossary

Calibration.
The match between a person's average confidence and their average accuracy; a gap in one direction is overconfidence and in the other underconfidence.
Cue utilization.
The view that judgments about one's own memory are inferences drawn from available cues rather than direct readings of memory strength, which is why cues unrelated to memory can sway them.
Feeling of knowing.
The sense that a currently unrecallable item is nonetheless stored in memory and could be recognised or retrieved later.
Feeling of rightness.
The sense of how correct a first, intuitive answer seems; when it is weak, people are more likely to reconsider and change the answer.
Judgment of learning.
A prediction, made during or after study, of how likely a studied item is to be remembered on a later test.
Meta-d′.
A signal detection measure of how well confidence ratings distinguish correct from incorrect decisions, expressed in the same units as the sensitivity of the decision itself.
Meta-emotion.
Feelings and beliefs about emotions, such as a parent's attitudes toward their own and their child's emotional states.
Metacognitive control.
The regulation of one's own cognition on the basis of monitoring, such as choosing what to restudy, when to stop searching memory or whether to change an answer.
Metacognitive efficiency.
Metacognitive sensitivity relative to task performance, usually the ratio meta-d′/d′; a value of 1 means confidence uses all the information available to the decision.
Metacognitive knowledge.
What a person knows or believes about thinking in general, about their own abilities and about the demands of particular tasks and strategies.
Metacognitive monitoring.
The ongoing assessment of one's own cognitive state, expressed in judgments such as confidence ratings, feelings of knowing and judgments of learning.
Metacognitive therapy.
A psychological treatment developed by Adrian Wells that targets people's beliefs about their own thinking, including beliefs about worry, in anxiety and depression.
Metamemory.
Knowledge about, and monitoring of, one's own memory.
Region of proximal learning.
A proposal about study-time allocation holding that learners' choices of what to study are guided by their judgments of learning and need not favour the hardest material.
Resolution.
The degree to which confidence is higher for one's correct answers than for one's errors, independent of the overall level of confidence; also called discrimination or sensitivity.
Self-regulated learning.
Learning in which the learner plans, monitors and adjusts their own study, drawing on metacognition, motivation and strategy use.
Type 2 judgment.
A judgment about one's own first-order decision or memory, such as a confidence rating given after an answer, as distinct from the first-order (type 1) decision itself.

Key Researchers

John Dunlosky

(Kent State University). Psychologist who studies how learners judge their own learning and how those judgments shape what and how they study. Faculty page · Wikidata

John H. Flavell

(Stanford University, 1928–2025). Developmental psychologist who introduced the term metacognition and studied how children come to monitor their own memory and understanding. Wikipedia · Wikidata

Stephen M. Fleming

(University College London). Cognitive neuroscientist who studies how confidence is computed in the brain and how metacognition can be measured relative to task performance. ORCID · Faculty page

Asher Koriat

(University of Haifa). Cognitive psychologist who proposed that feelings of knowing and judgments of learning are inferences drawn from cues rather than direct readings of memory. Faculty page · Wikidata

Hakwan Lau

(IBS Center for Neuroscience Imaging Research, Sungkyunkwan University). Cognitive neuroscientist who co-developed meta-d′, a signal-detection measure of how well confidence tracks accuracy. Faculty page · Google Scholar

Janet Metcalfe

(Columbia University). Cognitive psychologist who studies how judgments of learning guide study choices and who proposed the region of proximal learning model. Faculty page

Adrian Wells

(University of Manchester). Clinical psychologist who developed metacognitive therapy and a metacognitive model of worry and generalized anxiety. ORCID · Faculty page · Wikipedia · Wikidata

Frequently Asked Questions

What is metacognition?

Metacognition is thinking about one's own thinking: monitoring and regulating one's own memory, learning and decisions. Flavell described the new field and observed that young children are poor at monitoring their own memory and understanding (Flavell, 1979).

How do monitoring and control differ?

Monitoring is judging the state of one's own knowledge, as in rating how well a word pair has been learned, while control is acting on that judgment, as in choosing what to restudy. Experiments show that judgments of learning causally guide study choices, although the evidence was described as early (Metcalfe, 2009).

How do researchers measure metacognition?

They compare confidence ratings with accuracy across many trials. Measures based on signal detection theory are preferred over simple correlations because they are not distorted by a general tendency to give high or low ratings (Fleming & Lau, 2014).

Are people good judges of their own learning?

Often not. Judgments made while the answer is still in view overestimate how well it will be recalled later, when only the cue is available, an illusion of competence (Koriat & Bjork, 2005).

Does metacognition improve with age?

In a small cross sectional study of 56 people aged 11 to 41 using one perceptual task, metacognitive ability rose through adolescence and then levelled off in adulthood (Weil et al., 2013).

Do animals have metacognition?

Monkeys and a dolphin use an uncertain response on hard trials in a way that resembles human uncertainty monitoring, though whether this amounts to awareness of their own minds remains debated (Smith et al., 2003).

Can metacognitive skills be trained?

Meta-analyses of school interventions found that training in self-regulated learning improved performance, strategy use and motivation, with larger effects when researchers delivered it and in mathematics (Dignath & Büttner, 2008).

On the metacognitive account of worry, beliefs about one's own thinking keep anxiety going, and a meta-analysis of 25 studies found therapy targeting those beliefs effective, though the trials were few and small and allegiance bias was possible (Normann & Morina, 2018).

References

Ackerman, R., & Thompson, V. A. (2017). Meta-reasoning: Monitoring and control of thinking and reasoning. Trends in Cognitive Sciences, 21(8), 607-617. https://doi.org/10.1016/j.tics.2017.05.004

Ariel, R., Dunlosky, J., & Bailey, H. (2009). Agenda-based regulation of study-time allocation: When agendas override item-based monitoring. Journal of Experimental Psychology: General, 138(3), 432-447. https://doi.org/10.1037/a0015928

Bago, B., & De Neys, W. (2017). Fast logic?: Examining the time course assumption of dual process theory. Cognition, 158, 90-109. https://doi.org/10.1016/j.cognition.2016.10.014

Bjork, R. A., Dunlosky, J., & Kornell, N. (2013). Self-regulated learning: Beliefs, techniques, and illusions. Annual Review of Psychology, 64, 417-444. https://doi.org/10.1146/annurev-psych-113011-143823

Carruthers, P. (2008). Meta-cognition in animals: A skeptical look. Mind & Language, 23(1), 58-89. https://doi.org/10.1111/j.1468-0017.2007.00329.x

Dignath, C., & Büttner, G. (2008). Components of fostering self-regulated learning among students. A meta-analysis on intervention studies at primary and secondary school level. Metacognition and Learning, 3(3), 231-264. https://doi.org/10.1007/s11409-008-9029-x

Dunlosky, J., & Rawson, K. A. (2012). Overconfidence produces underachievement: Inaccurate self evaluations undermine students' learning and retention. Learning and Instruction, 22(4), 271-280. https://doi.org/10.1016/j.learninstruc.2011.08.003

Efklides, A. (2011). Interactions of metacognition with motivation and affect in self-regulated learning: The MASRL model. Educational Psychologist, 46(1), 6-25. https://doi.org/10.1080/00461520.2011.538645

Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry. American Psychologist, 34(10), 906–911. https://doi.org/10.1037/0003-066X.34.10.906

Fleming, S. M. (2024). Metacognition and confidence: A review and synthesis. Annual Review of Psychology, 75, 241-268. https://doi.org/10.1146/annurev-psych-022423-032425

Fleming, S. M., & Dolan, R. J. (2012). The neural basis of metacognitive ability. Philosophical Transactions of the Royal Society B: Biological Sciences, 367(1594), 1338-1349. https://doi.org/10.1098/rstb.2011.0417

Fleming, S. M., & Lau, H. C. (2014). How to measure metacognition. Frontiers in Human Neuroscience, 8, 443. https://doi.org/10.3389/fnhum.2014.00443

Fleming, S. M., Weil, R. S., Nagy, Z., Dolan, R. J., & Rees, G. (2010). Relating introspective accuracy to individual differences in brain structure. Science, 329(5998), 1541-1543. https://doi.org/10.1126/science.1191883

Frith, C. D. (2012). The role of metacognition in human social interactions. Philosophical Transactions of the Royal Society B: Biological Sciences, 367(1599), 2213-2223. https://doi.org/10.1098/rstb.2012.0123

Gignac, G. E., & Zajenkowski, M. (2020). The Dunning-Kruger effect is (mostly) a statistical artefact: Valid approaches to testing the hypothesis with individual differences data. Intelligence, 80, 101449. https://doi.org/10.1016/j.intell.2020.101449

Gottman, J. M., Katz, L. F., & Hooven, C. (1996). Parental meta-emotion philosophy and the emotional life of families: Theoretical models and preliminary data. Journal of Family Psychology, 10(3), 243-268. https://doi.org/10.1037/0893-3200.10.3.243

Guggenmos, M. (2021). Measuring metacognitive performance: Type 1 performance dependence and test-retest reliability. Neuroscience of Consciousness, 2021(1), niab040. https://doi.org/10.1093/nc/niab040

Hart, J. T. (1965). Memory and the feeling-of-knowing experience. Journal of Educational Psychology, 56(4), 208-216. https://doi.org/10.1037/h0022263

Hattie, J., Biggs, J., & Purdie, N. (1996). Effects of learning skills interventions on student learning: A meta-analysis. Review of Educational Research, 66(2), 99-136. https://doi.org/10.3102/00346543066002099

Hoven, M., Lebreton, M., Engelmann, J. B., Denys, D., Luigjes, J., & van Holst, R. J. (2019). Abnormalities of confidence in psychiatry: An overview and future perspectives. Translational Psychiatry, 9(1), 268. https://doi.org/10.1038/s41398-019-0602-7

Kepecs, A., Uchida, N., Zariwala, H. A., & Mainen, Z. F. (2008). Neural correlates, computation and behavioural impact of decision confidence. Nature, 455(7210), 227-231. https://doi.org/10.1038/nature07200

Koriat, A. (1993). How do we know that we know? The accessibility model of the feeling of knowing. Psychological Review, 100(4), 609-639. https://doi.org/10.1037/0033-295X.100.4.609

Koriat, A. (1997). Monitoring one's own knowledge during study: A cue-utilization approach to judgments of learning. Journal of Experimental Psychology: General, 126(4), 349-370. https://doi.org/10.1037/0096-3445.126.4.349

Koriat, A., & Bjork, R. A. (2005). Illusions of competence in monitoring one's knowledge during study. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31(2), 187-194. https://doi.org/10.1037/0278-7393.31.2.187

Koriat, A., & Levy-Sadot, R. (2001). The combined contributions of the cue-familiarity and accessibility heuristics to feelings of knowing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27(1), 34-53. https://doi.org/10.1037/0278-7393.27.1.34

Kornell, N., & Bjork, R. A. (2008). Learning concepts and categories: Is spacing the "enemy of induction"? Psychological Science, 19(6), 585-592. https://doi.org/10.1111/j.1467-9280.2008.02127.x

Kornell, N., Son, L. K., & Terrace, H. S. (2007). Transfer of metacognitive skills and hint seeking in monkeys. Psychological Science, 18(1), 64-71. https://doi.org/10.1111/j.1467-9280.2007.01850.x

Kruger, J., & Dunning, D. (1999). Unskilled and unaware of it: How difficulties in recognizing one's own incompetence lead to inflated self-assessments. Journal of Personality and Social Psychology, 77(6), 1121-1134. https://doi.org/10.1037/0022-3514.77.6.1121

Lopez-Morinigo, J.-D., Ajnakina, O., Martínez, A. S.-E., Escobedo-Aedo, P.-J., Ruiz-Ruano, V. G., Sánchez-Alonso, S., Mata-Iturralde, L., Muñoz-Lorenzo, L., Ochoa, S., Baca-García, E., & David, A. S. (2020). Can metacognitive interventions improve insight in schizophrenia spectrum disorders? A systematic review and meta-analysis. Psychological Medicine, 50(14), 2289–2301. https://doi.org/10.1017/S0033291720003384

Maniscalco, B., & Lau, H. (2012). A signal detection theoretic approach for estimating metacognitive sensitivity from confidence ratings. Consciousness and Cognition, 21(1), 422-430. https://doi.org/10.1016/j.concog.2011.09.021

Mazancieux, A., Fleming, S. M., Souchay, C., & Moulin, C. J. A. (2020). Is there a G factor for metacognition? Correlations in retrospective metacognitive sensitivity across tasks. Journal of Experimental Psychology: General, 149(9), 1788-1799. https://doi.org/10.1037/xge0000746

Metcalfe, J. (2009). Metacognitive judgments and control of study. Current Directions in Psychological Science, 18(3), 159-163. https://doi.org/10.1111/j.1467-8721.2009.01628.x

Metcalfe, J., & Kornell, N. (2005). A region of proximal learning model of study time allocation. Journal of Memory and Language, 52(4), 463-477. https://doi.org/10.1016/j.jml.2004.12.001

Metcalfe, J., Schwartz, B. L., & Joaquim, S. G. (1993). The cue-familiarity heuristic in metacognition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19(4), 851-861. https://doi.org/10.1037/0278-7393.19.4.851

Moritz, S., & Woodward, T. S. (2007). Metacognitive training in schizophrenia: From basic research to knowledge translation and intervention. Current Opinion in Psychiatry, 20(6), 619-625. https://doi.org/10.1097/YCO.0b013e3282f0b8ed

Mueller, M. L., Dunlosky, J., Tauber, S. K., & Rhodes, M. G. (2014). The font-size effect on judgments of learning: Does it exemplify fluency effects or reflect people's beliefs about memory? Journal of Memory and Language, 70, 1-12. https://doi.org/10.1016/j.jml.2013.09.007

Nelson, T. O. (1984). A comparison of current measures of the accuracy of feeling-of-knowing predictions. Psychological Bulletin, 95(1), 109-133. https://doi.org/10.1037/0033-2909.95.1.109

Nelson, T. O., & Dunlosky, J. (1991). When people's judgments of learning (JOLs) are extremely accurate at predicting subsequent recall: The "delayed-JOL effect." Psychological Science, 2(4), 267-271. https://doi.org/10.1111/j.1467-9280.1991.tb00147.x

Nelson, T. O., & Narens, L. (1990). Metamemory: A theoretical framework and new findings. Psychology of Learning and Motivation, 26, 125-173. https://doi.org/10.1016/S0079-7421(08)60053-5

Normann, N., & Morina, N. (2018). The efficacy of metacognitive therapy: A systematic review and meta-analysis. Frontiers in Psychology, 9, 2211. https://doi.org/10.3389/fpsyg.2018.02211

Rahnev, D., Desender, K., Lee, A. L. F., Adler, W. T., Aguilar-Lleyda, D., Akdoğan, B., Arbuzova, P., Atlas, L. Y., Balcı, F., Bang, J. W., Bègue, I., Birney, D. P., Brady, T. F., Calder-Travis, J., Chetverikov, A., Clark, T. K., Davranche, K., Denison, R. N., Dildine, T. C., . . . Zylberberg, A. (2020). The Confidence Database. Nature Human Behaviour, 4(3), 317-325. https://doi.org/10.1038/s41562-019-0813-1

Reder, L. M., & Ritter, F. E. (1992). What determines initial feeling of knowing? Familiarity with question terms, not with the answer. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18(3), 435-451. https://doi.org/10.1037/0278-7393.18.3.435

Rhodes, M. G., & Castel, A. D. (2008). Memory predictions are influenced by perceptual information: Evidence for metacognitive illusions. Journal of Experimental Psychology: General, 137(4), 615-625. https://doi.org/10.1037/a0013684

Rouault, M., McWilliams, A., Allen, M. G., & Fleming, S. M. (2018). Human metacognition across domains: Insights from individual differences and neuroimaging. Personality Neuroscience, 1, e17. https://doi.org/10.1017/pen.2018.16

Rouault, M., Seow, T. X. F., Gillan, C. M., & Fleming, S. M. (2018). Psychiatric symptom dimensions are associated with dissociable shifts in metacognition but not task performance. Biological Psychiatry, 84(6), 443-451. https://doi.org/10.1016/j.biopsych.2017.12.017

Schraw, G., & Dennison, R. S. (1994). Assessing metacognitive awareness. Contemporary Educational Psychology, 19(4), 460-475. https://doi.org/10.1006/ceps.1994.1033

Seow, T. X. F., Rouault, M., Gillan, C. M., & Fleming, S. M. (2021). How local and global metacognition shape mental health. Biological Psychiatry, 90(7), 436-446. https://doi.org/10.1016/j.biopsych.2021.05.013

Shea, N., Boldt, A., Bang, D., Yeung, N., Heyes, C., & Frith, C. D. (2014). Supra-personal cognitive control and metacognition. Trends in Cognitive Sciences, 18(4), 186-193. https://doi.org/10.1016/j.tics.2014.01.006

Smith, J. D., Shields, W. E., & Washburn, D. A. (2003). The comparative psychology of uncertainty monitoring and metacognition. Behavioral and Brain Sciences, 26(3), 317-339. https://doi.org/10.1017/S0140525X03000086

Son, L. K., & Metcalfe, J. (2000). Metacognitive and control strategies in study-time allocation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26(1), 204-221. https://doi.org/10.1037/0278-7393.26.1.204

Thompson, V. A., Prowse Turner, J. A., & Pennycook, G. (2011). Intuition, reason, and metacognition. Cognitive Psychology, 63(3), 107-140. https://doi.org/10.1016/j.cogpsych.2011.06.001

Vaccaro, A. G., & Fleming, S. M. (2018). Thinking about thinking: A coordinate-based meta-analysis of neuroimaging studies of metacognitive judgements. Brain and Neuroscience Advances, 2, Article 2398212818810591. https://doi.org/10.1177/2398212818810591

Veenman, M. V. J., Van Hout-Wolters, B. H. A. M., & Afflerbach, P. (2006). Metacognition and learning: Conceptual and methodological considerations. Metacognition and Learning, 1(1), 3-14. https://doi.org/10.1007/s11409-006-6893-0

Weil, L. G., Fleming, S. M., Dumontheil, I., Kilford, E. J., Weil, R. S., Rees, G., Dolan, R. J., & Blakemore, S.-J. (2013). The development of metacognitive ability in adolescence. Consciousness and Cognition, 22(1), 264-271. https://doi.org/10.1016/j.concog.2013.01.004

Wells, A. (1995). Meta-cognition and worry: A cognitive model of generalized anxiety disorder. Behavioural and Cognitive Psychotherapy, 23(3), 301-320. https://doi.org/10.1017/S1352465800015897

Wells, A., & Cartwright-Hatton, S. (2004). A short form of the metacognitions questionnaire: Properties of the MCQ-30. Behaviour Research and Therapy, 42(4), 385-396. https://doi.org/10.1016/S0005-7967(03)00147-5

Zimmerman, B. J. (2002). Becoming a self-regulated learner: An overview. Theory Into Practice, 41(2), 64-70. https://doi.org/10.1207/s15430421tip4102_2