Hallucinations

Abstract

A hallucination is a perception that arises without any external stimulus yet is experienced with the full conviction of a real percept. Jean-Étienne Dominique Esquirol fixed the modern definition in the early nineteenth century, separating a percept without an object from an illusion, a real object misperceived. Two cognitive accounts dominate: a source-monitoring failure, in which self-generated mental events are misattributed to an external source, and a predictive-coding account, in which perception is dominated by prior expectation. Hallucinations are not confined to illness, occurring in a substantial minority of healthy people, which places them on a continuum rather than marking a categorical break. This article defines the hallucination, surveys its modalities and types, sets out the two mechanisms, reviews the continuum and the neuroimaging of the hallucinating brain, and describes how hallucinations are measured, with interactive demonstrations of each mechanism.

Keywords: hallucination, source monitoring, predictive coding

What a Hallucination Is

A hallucination is a sensory experience that occurs in the absence of a corresponding external stimulus but is accompanied by the compelling sense that the perceived object or event is real. The classical formulation is Esquirol's: a person who has “a thorough conviction of perceiving a sensation for which there is no external object” is hallucinating. The force of the definition lies in two clauses held together — there is no stimulus, and yet the experience has the qualities and the credibility of genuine perception. A hallucination is therefore not a vague impression or a thought experienced as a thought; it is a percept, located in external space, carrying the vividness and involuntariness that ordinarily belong only to perceptions of real things.

This is what separates a hallucination from an illusion. An illusion is a misperception of something that is actually present — a coat on a door mistaken for an intruder, a stick in water seen as bent. A hallucination has no external object at all; the coat and the stick are simply not there. The distinction, which Esquirol drew explicitly, remains the organizing contrast of the field: illusions distort the input, hallucinations manufacture it. A hallucination is also distinct from a delusion, which is a false belief rather than a false percept, though the two frequently travel together when a person tries to make sense of an anomalous experience.

Types of Hallucinations

In the Medical Subject Headings (MeSH) vocabulary, Hallucinations is classed as a type of perceptual disorder, and the descriptor carries one narrower descriptor of its own that also has a page on this site. MeSH is an indexing classification built for retrieving the literature, not a theory of perception, so its single formal subtype should be read as one curated distinction among several ways hallucinations are divided — by modality, by content, and by clinical context — rather than as the field's complete taxonomy. The axes are largely orthogonal: a hallucination has a modality (what sense it occupies), a content type, and a context (who experiences it and why), and these vary independently of one another.

Charles Bonnet syndrome is instructive precisely because it is the uncomplicated case: the hallucinations are vivid and often elaborate, yet the person retains full insight that they are not real, and there is no psychiatric illness driving them — only the deafferented visual system filling the void left by lost input. It shows that a hallucination need not imply madness, and it anticipates the continuum theme developed below. The broader divisions MeSH does not formalize here are taken up in the next section: the modality axis, which distinguishes auditory, visual, olfactory, gustatory, tactile, and other forms.

Modalities

Hallucinations can occupy any sense, and the modality shapes both the typical content and the diagnostic implication. Auditory hallucinations are the most frequent in psychosis and the most intensively researched; their paradigm case is the auditory verbal hallucination (AVH), the experience of hearing one or more voices speaking when no one is present. Voices may comment on the person's actions, converse with one another, or issue commands, and they are typically experienced as originating in external space with an identity and a will of their own. Visual hallucinations range from simple flashes and geometric forms to fully formed figures and scenes; they are comparatively more common in organic and neurological conditions — delirium, Lewy body dementia, eye disease, and drug states — than in schizophrenia, a difference that carries diagnostic weight.

The remaining modalities are less common but clinically significant. Olfactory and gustatory hallucinations — phantom smells and tastes — often point to temporal-lobe epilepsy or other organic causes and can form the aura of a seizure. Tactile (or somatic) hallucinations include the sensation of insects crawling on or under the skin, characteristic of stimulant intoxication and withdrawal. A separate category, hypnagogic and hypnopompic hallucinations, occur in the transition into and out of sleep and are experienced by many healthy people; their prevalence in the general population is one of the clearest pieces of evidence that a hallucination is not in itself a sign of disease (Johns & van Os, 2001).

The modality profile is itself informative. Auditory-predominant hallucinations with preserved consciousness are more suggestive of a primary psychotic disorder, whereas prominent visual hallucinations, especially with clouding of consciousness, shift the suspicion toward an organic or toxic cause. This diagnostic use of modality rests on large phenomenological surveys that catalogue how the different forms actually present across conditions (McCarthy-Jones et al., 2014).

The Source-Monitoring Account

The first of the two dominant cognitive theories reframes the hallucination as an error of attribution rather than of sensation. Source monitoring is the ordinary process by which we judge where a mental event came from — whether a remembered sentence was spoken aloud or only imagined, whether an image is a memory or a fantasy. Richard Bentall's influential 1990 review argued that hallucinations arise when this process fails: a self-generated mental event, such as inner speech or a vivid image, is misattributed to an external source and so is experienced as a perception rather than as one's own thought (Bentall, 1990). On this account the voice a person hears is, in some sense, their own inner speech, mislabelled as coming from outside.

The attraction of the source-monitoring account is that it connects hallucinations to a normal cognitive faculty that can be studied in anyone. Experiments show that people prone to hallucinations are more likely, under uncertainty, to judge self-generated material as externally presented — a bias toward “externalizing” that is measurable with signal-detection methods and that tracks hallucination-proneness across clinical and healthy groups. The theory also explains why AVHs so often take the form of speech: inner speech is the self-generated mental event most naturally misattributed as an external voice, which is why the content is verbal and why neuroimaging of AVHs implicates the brain's speech and language systems (see below). A closely related mechanism specifies how the misattribution happens at the neural level: the self-monitoring or corollary-discharge account holds that whenever the brain generates an action — including inner speech — it issues an efference copy that predicts the sensory consequences and tags them as self-produced; if that corollary discharge fails, self-generated speech arrives untagged and is experienced as an external voice (Ford & Mathalon, 2005). This is the forward-model counterpart of source monitoring, and it is the bridge to the predictive-coding account that follows, since a corollary discharge is precisely a top-down prediction. The framework has since been integrated with others into a broader cognitive model of auditory hallucinations spanning clinical and nonclinical populations (Waters et al., 2012).

The Predictive-Coding Account

The second account comes from the predictive-coding view of perception, in which the brain is not a passive receiver of sensory data but an active prediction machine that continually generates expectations about the world and updates them against incoming evidence. Perception, on this view, is the brain's best guess — a balance struck between top-down priors (what it expects) and bottom-up input (what the senses report). A hallucination results when that balance tips too far toward the prior: expectation comes to dominate evidence, and the brain perceives what it predicts rather than what is there. Philip Corlett and colleagues formalized this as the “strong priors” account, in which hallucinations are perception overweighted by prior expectation (Corlett et al., 2019).

The decisive support is experimental. In a conditioning paradigm, Powers, Mathys, and Corlett paired a tone with a visual stimulus until participants came to “hear” the tone when only the light was presented — an induced hallucination — and showed that this susceptibility was greatest in people who hallucinate in daily life, and that it reflected a measurable overweighting of priors in a computational model of their behaviour (Powers et al., 2017). The same laboratory had earlier framed hallucinations in general as top-down effects on perception, locating the fault not in the sensory input but in the weight given to expectation (Powers et al., 2016). The predictive-coding account and the source-monitoring account are not rivals so much as descriptions at different levels: a mislabelled self-generated event is one way a strong prior can be realized, and both locate the hallucination downstream of intact sensory organs, in the inferential machinery that constructs a percept.

The Continuum Across Populations

One of the most consequential findings of the last three decades is that hallucinations are not confined to psychiatric illness. Large general-population surveys find that a substantial minority of people — commonly estimated in the range of five to fifteen percent for lifetime auditory or visual hallucinatory experiences — report having hallucinated while never meeting criteria for a psychotic disorder, a pattern that supports a continuity between psychotic experiences and normal functioning rather than a categorical divide (Johns & van Os, 2001). The experience, in other words, is far more common than the illness.

The point is sharpened by direct study of healthy voice-hearers. Sommer and colleagues assembled a sample of individuals who hear voices regularly yet are psychiatrically well, and found that their hallucinations were phenomenologically similar to those in schizophrenia in form — but differed in content and context: the voices were more often neutral or positive, the person retained control and insight, and the onset was not marked by the distress and loss of function that define the clinical case (Sommer et al., 2010). A systematic comparison across diagnostic classes confirms that hallucinations share a common phenomenological core wherever they occur, while the features that predict clinical need — distress, loss of control, negative content — vary across groups (Waters & Fernyhough, 2017).

The continuum reframes the central question. If hallucinating is something many well people do, then the explanatory target is not merely why hallucinations occur but why they become distressing and disabling in some people and not others — a shift of emphasis from the presence of the experience to the appraisal of it, with direct consequences for treatment.

The Hallucinating Brain

Neuroimaging has made it possible to watch the brain while a person hallucinates, and the results anchor the cognitive theories in neural activity. The landmark demonstration for vision came from Dominic ffytche and colleagues, who scanned people with Charles Bonnet syndrome during their visual hallucinations and found that the content of a hallucination tracked activity in the corresponding specialized region of visual cortex: hallucinations of colour engaged colour areas, of faces the face area, of text the word-form area (ffytche et al., 1998). The hallucination, that is, recruited the very cortex that would process the real thing — strong evidence that it is a genuine perceptual event, not merely imagined.

For auditory verbal hallucinations, functional imaging shows activation in the brain's speech-production and speech-perception systems during the experience of hearing voices, consistent with the source-monitoring idea that the voice is self-generated inner speech experienced as external (Shergill et al., 2000). Quantitative syntheses across many studies have identified a distributed network — language, auditory, and salience regions among them — reliably engaged during hallucinations (Zmigrod et al., 2016), and broad reviews of the structural and functional imaging literature converge on the same picture of hallucinations as the activity of perceptual systems operating without their normal sensory drive (Allen et al., 2008). A general model of complex visual hallucinations ties these findings together by proposing that they arise from a specific combination of impaired attentional and perceptual processing (Collerton et al., 2005).

Assessment and Measurement

Because a hallucination is a private experience, it can only be measured through report, and the central methodological problem is to capture its many dimensions reliably. Early assessment asked simply whether hallucinations were present or absent. Modern instruments instead treat the hallucination as multidimensional, rating its frequency, duration, loudness, location (inside or outside the head), the degree of negative content, the distress it causes, and the person's sense of control and conviction. This multidimensional approach is what makes the continuum visible: it is precisely on the distress-and-control dimensions, not on the presence of the voice, that healthy and clinical voice-hearers diverge.

Structured phenomenological surveys have been used to map the landscape of hallucinatory experience in large samples, revealing subtypes and regularities that a present-or-absent question would miss (McCarthy-Jones et al., 2014). Alongside clinical rating scales, experimental tasks give objective, behavioural measures of hallucination-proneness: signal-detection paradigms quantify the bias to hear a signal in noise, and conditioning paradigms induce hallucination-like percepts under controlled conditions and fit computational models to the result (Powers et al., 2017). The comprehensive reference for the measurement and science of hallucinations across all these approaches is the book-length treatment by Aleman and Larøi (Aleman & Larøi, 2008), and the integrated cognitive model of Waters and colleagues provides the framework that ties the behavioural measures to mechanism (Waters et al., 2012; Larøi et al., 2012).

Figure

Figure 1

A Perception Constructed From Prior and Evidence

Interactive Demonstrations

The three demonstrations below make the core ideas manipulable. The first models source monitoring as a signal-detection problem, letting the reader move the decision criterion and watch how an externalizing bias turns self-generated events into “heard” voices. The second shows the predictive-coding balance directly, letting the reader set the strength of the prior against the clarity of the evidence and see when the percept detaches from the input. The third plots the clinical–nonclinical continuum, showing how the same hallucinatory experience separates into distressing and non-distressing cases along the dimensions of control and content rather than presence.

Worked Example

Consider the source-monitoring account cast as a signal-detection problem. A listener must decide, on each trial, whether a faint voice is present in a burst of noise. Over a session the experimenter records how often the listener says “voice” when a voice was truly presented (a hit) and how often they say “voice” when none was (a false alarm). Suppose two listeners are tested on 100 signal-absent trials and 100 signal-present trials:

- Listener A (low hallucination-proneness): 70 hits, 10 false alarms. - Listener B (high hallucination-proneness): 72 hits, 40 false alarms.

The two have almost identical hit rates — their raw sensitivity to a real voice is much the same. What separates them is the false-alarm rate: Listener B says “voice” to empty noise four times as often. In signal-detection terms, B's sensitivity (the separation of signal from noise) is barely different, but B's criterion for reporting a voice is far more liberal. Converting the false-alarm rates to a rough bias measure, A requires strong evidence before reporting a voice while B will report one on thin evidence — an “externalizing” bias, the tendency to assign ambiguous internal events to an external source.

This is exactly the pattern the source-monitoring account predicts: hallucination-proneness shows up not as better or worse hearing but as a shifted decision, a willingness to perceive a voice where there is none. It also dovetails with the predictive-coding reading — a liberal criterion is what a strong prior “expect a voice” looks like when expressed as a decision rule — and it shows why hallucinations can be studied quantitatively in people who are entirely well: the externalizing bias is a graded trait, measurable in anyone, not a symptom that is simply present or absent.

Discussion

The hallucination is valuable to cognitive science out of proportion to how strange it first appears, because it exposes the constructed nature of ordinary perception. In everyday life the seamless fit between what we perceive and what is there makes it easy to believe that perception simply registers the world. The hallucination breaks that assumption open: here is a full-blooded percept, with all the vividness and conviction of the real thing, built entirely from within. Whatever machinery can do that in the absence of a stimulus must be the same machinery that, in the presence of one, constructs the percept we take for granted — which is why both leading theories treat hallucination not as an alien intrusion but as normal perception running under abnormal conditions.

The two accounts answer complementary questions. Source monitoring explains the attribution — why a self-generated event is experienced as external, and why the auditory-verbal form is so common. Predictive coding explains the construction — how expectation can override evidence to generate a percept at all, and why hallucination-proneness tracks a measurable overweighting of priors. Binding them is the continuum: once it is clear that many well people hallucinate, the field's question shifts from the presence of the experience to its appraisal, and the clinically urgent variables become distress, control, and content rather than the bare fact of the voice. That reframing, more than any single mechanism, is the practical legacy of the last thirty years of hallucination research.

Current Directions

Contemporary work is consolidating the computational turn and pressing the continuum toward the clinic. The strongest recent thread is the computational-psychiatry program built on predictive coding: having shown that hallucination-like percepts can be induced by conditioning and that susceptibility reflects an overweighting of priors fitted in a formal model, the field is now using such models to ask whether the balance of prior and evidence can index who will hallucinate and who will be distressed by it (Powers et al., 2017; Corlett et al., 2019). A second thread develops the integrated cognitive model of auditory hallucinations, seeking the common mechanism that spans schizophrenia, healthy voice-hearers, and the hallucinations of neurological and sensory disease, and asking what the shared phenomenological core and the diverging clinical features imply for a unified account (Waters et al., 2012; Waters & Fernyhough, 2017). A third keeps refining measurement, using large phenomenological surveys and quantitative neuroimaging syntheses to map subtypes and their neural correlates with increasing precision (McCarthy-Jones et al., 2014; Zmigrod et al., 2016). The open questions are no longer whether hallucinations can be studied scientifically but how finely the attributional and inferential components can be separated, and whether a computational marker of prior-overweighting can guide intervention before distress sets in.

Common Misconceptions

Only people with a mental illness hallucinate.

A substantial minority of the healthy population report hallucinatory experiences without ever meeting criteria for a psychotic disorder, which places hallucinations on a continuum rather than marking a categorical sign of illness (Johns & van Os, 2001; Sommer et al., 2010).

A hallucination is the same as an illusion.

An illusion is a real object misperceived; a hallucination has no external object at all. Esquirol drew exactly this distinction, and it remains the organizing contrast of the field (Aleman & Larøi, 2008).

Hallucinations are imagined, not really perceived.

Neuroimaging shows that a hallucination recruits the same specialized sensory cortex that would process the real stimulus, so it is a genuine perceptual event rather than a vivid thought (ffytche et al., 1998; Shergill et al., 2000).

Hearing voices always means schizophrenia.

Auditory verbal hallucinations occur across many conditions and in well people; what predicts clinical need is the distress, loss of control, and negative content, not the voice itself (Waters & Fernyhough, 2017).

Glossary

Key Researchers

Richard P. Bentall

(University of Sheffield). Clinical psychologist whose 1990 review reframed hallucinations as a failure of source monitoring — the misattribution of self-generated mental events to an external source — a cornerstone of the psychological account. ORCID · Faculty

Philip R. Corlett

(Yale University). Associate professor of psychiatry whose strong-priors predictive-coding account frames hallucinations as perception dominated by prior expectation, tested in conditioning experiments that induce hallucination-like percepts. Faculty · Wikidata

Jean-Étienne Dominique Esquirol

(Salpêtrière, Paris). French psychiatrist who gave the modern definition of the hallucination in the early nineteenth century, distinguishing a percept without an object from an illusion, a real object misperceived. Wikipedia

Dominic H. ffytche

(King's College London). Professor of visual psychiatry whose fMRI work tied the content of a visual hallucination to activity in the corresponding specialized visual cortex, and the leading authority on Charles Bonnet syndrome. ORCID · Faculty

Flavie Waters

(University of Western Australia). Clinical neuroscientist who led the integrated cognitive model of auditory hallucinations and the cross-diagnostic review showing hallucinations span clinical and nonclinical populations. ORCID · Faculty

Frequently Asked Questions

What is a hallucination?

A hallucination is a perception that occurs without any external stimulus but is experienced with the full conviction and vividness of a real percept. It can occur in any sense, and it is located in external space with the involuntariness that ordinarily belongs only to perceptions of real things.

How is a hallucination different from an illusion?

An illusion is a misperception of something that is actually present, such as a coat mistaken for an intruder. A hallucination has no external object at all. Esquirol drew this distinction in the nineteenth century, and it remains the organizing contrast of the field.

Does hallucinating mean a person has a mental illness?

No. A substantial minority of healthy people experience hallucinations without ever developing a psychotic disorder. What distinguishes the clinical cases is not the presence of the experience but the distress it causes, the loss of control over it, and its negative content.

Why are voices the most common kind?

Auditory verbal hallucinations predominate in psychosis, and the source-monitoring account explains why: inner speech is the self-generated mental event most readily misattributed as an external voice, which is also why neuroimaging of voices engages the brain's speech systems.

What causes hallucinations according to cognitive science?

Two accounts dominate. Source monitoring holds that self-generated mental events are misattributed to an external source. Predictive coding holds that perception is a balance of prior expectation and sensory evidence, and that a hallucination results when a strong prior overrides the evidence.

Are hallucinations real perceptions or just imagination?

They are genuine perceptual events. Brain imaging shows that a hallucination activates the same specialized sensory cortex that would process the corresponding real stimulus (colour areas for colour, face areas for faces), which is why it carries the force of a real percept.

What is Charles Bonnet syndrome?

It is the occurrence of complex visual hallucinations in people with significant vision loss who have intact cognition and full insight that the images are not real. It shows that a hallucination can arise from a deafferented sensory system without any psychiatric illness.

How are hallucinations measured?

Through multidimensional report and experimental tasks. Clinical scales rate frequency, loudness, location, control, distress, and content; signal-detection and conditioning paradigms give objective, behavioural measures of the bias toward perceiving a signal that is not there.

References

Aleman, A., & Larøi, F. (2008). Hallucinations: The science of idiosyncratic perception. American Psychological Association. https://doi.org/10.1037/11751-000

Allen, P., Larøi, F., McGuire, P. K., & Aleman, A. (2008). The hallucinating brain: A review of structural and functional neuroimaging studies of hallucinations. Neuroscience & Biobehavioral Reviews, 32(1), 175–191. https://doi.org/10.1016/j.neubiorev.2007.07.012

Bentall, R. P. (1990). The illusion of reality: A review and integration of psychological research on hallucinations. Psychological Bulletin, 107(1), 82–95. https://doi.org/10.1037/0033-2909.107.1.82

Collerton, D., Perry, E., & McKeith, I. (2005). Why people see things that are not there: A novel Perception and Attention Deficit model for recurrent complex visual hallucinations. Behavioral and Brain Sciences, 28(6), 737–757. https://doi.org/10.1017/S0140525X05000130

Corlett, P. R., Horga, G., Fletcher, P. C., Alderson-Day, B., Schmack, K., & Powers, A. R. (2019). Hallucinations and strong priors. Trends in Cognitive Sciences, 23(2), 114–127. https://doi.org/10.1016/j.tics.2018.12.001

ffytche, D. H., Howard, R. J., Brammer, M. J., David, A., Woodruff, P., & Williams, S. (1998). The anatomy of conscious vision: An fMRI study of visual hallucinations. Nature Neuroscience, 1(8), 738–742. https://doi.org/10.1038/3738

Ford, J. M., & Mathalon, D. H. (2005). Corollary discharge dysfunction in schizophrenia: Can it explain auditory hallucinations? International Journal of Psychophysiology, 58(2-3), 179–189. https://doi.org/10.1016/j.ijpsycho.2005.01.014

Johns, L. C., & van Os, J. (2001). The continuity of psychotic experiences in the general population. Clinical Psychology Review, 21(8), 1125–1141. https://doi.org/10.1016/S0272-7358(01)00103-9

Larøi, F., Sommer, I. E., Blom, J. D., Fernyhough, C., ffytche, D. H., Hugdahl, K., Johns, L. C., McCarthy-Jones, S., Preti, A., Raballo, A., Slotema, C. W., Stephane, M., & Waters, F. (2012). The characteristic features of auditory verbal hallucinations in clinical and nonclinical groups: State-of-the-art overview and future directions. Schizophrenia Bulletin, 38(4), 724–733. https://doi.org/10.1093/schbul/sbs061

McCarthy-Jones, S., Trauer, T., Mackinnon, A., Sims, E., Thomas, N., & Copolov, D. L. (2014). A new phenomenological survey of auditory hallucinations: Evidence for subtypes and implications for theory and practice. Schizophrenia Bulletin, 40(1), 231–235. https://doi.org/10.1093/schbul/sbs156

Powers, A. R., Kelley, M., & Corlett, P. R. (2016). Hallucinations as top-down effects on perception. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, 1(5), 393–400. https://doi.org/10.1016/j.bpsc.2016.04.003

Powers, A. R., Mathys, C., & Corlett, P. R. (2017). Pavlovian conditioning–induced hallucinations result from overweighting of perceptual priors. Science, 357(6351), 596–600. https://doi.org/10.1126/science.aan3458

Shergill, S. S., Brammer, M. J., Williams, S. C. R., Murray, R. M., & McGuire, P. K. (2000). Mapping auditory hallucinations in schizophrenia using functional magnetic resonance imaging. Archives of General Psychiatry, 57(11), 1033–1038. https://doi.org/10.1001/archpsyc.57.11.1033

Sommer, I. E. C., Daalman, K., Rietkerk, T., Diederen, K. M., Bakker, S., Wijkstra, J., & Boks, M. P. M. (2010). Healthy individuals with auditory verbal hallucinations; who are they? Psychiatric assessments of a selected sample of 103 subjects. Schizophrenia Bulletin, 36(3), 633–641. https://doi.org/10.1093/schbul/sbn130

Waters, F., Allen, P., Aleman, A., Fernyhough, C., Woodward, T. S., Badcock, J. C., Barkus, E., Johns, L., Varese, F., Menon, M., Vercammen, A., & Larøi, F. (2012). Auditory hallucinations in schizophrenia and nonschizophrenia populations: A review and integrated model of cognitive mechanisms. Schizophrenia Bulletin, 38(4), 683–692. https://doi.org/10.1093/schbul/sbs045

Waters, F., & Fernyhough, C. (2017). Hallucinations: A systematic review of points of similarity and difference across diagnostic classes. Schizophrenia Bulletin, 43(1), 32–43. https://doi.org/10.1093/schbul/sbw132

Zmigrod, L., Garrison, J. R., Carr, J., & Simons, J. S. (2016). The neural mechanisms of hallucinations: A quantitative meta-analysis of neuroimaging studies. Neuroscience & Biobehavioral Reviews, 69, 113–123. https://doi.org/10.1016/j.neubiorev.2016.05.037