Levels of Processing

The levels of processing (LOP) framework, introduced by Fergus Craik and Robert Lockhart in their paper "Levels of Processing: A Framework for Memory Research" (Craik & Lockhart, 1972), fundamentally redirected memory research by arguing that how information is processed during encoding — not where it is stored — determines how well it will be remembered. This framework challenged the then-dominant Atkinson and Shiffrin (1968) multi-store model and remains one of the most cited and influential ideas in the history of cognitive psychology.

Key Structures

• Left inferior prefrontal cortex — The left-lateralized prefrontal region (Brodmann areas 44, 45, 47) consistently associated with semantic processing, controlled retrieval, and the deep encoding operations that produce strong memories.
• Hippocampus — A medial temporal lobe structure essential for binding the products of deep processing into coherent episodic memory traces. Hippocampal activation during encoding predicts subsequent memory success.
• Temporal cortex — The lateral temporal lobe regions that store semantic knowledge and are activated during meaning-based processing, providing the conceptual content for deep encoding.

Key Functions

Demonstrate that deeper, more meaningful processing of information during encoding leads to stronger, more durable memory traces, shifting the theoretical focus from structural models of memory stores to the cognitive operations performed on incoming information.

Historical Context: Challenging the Multi-Store Model

By the early 1970s, the Atkinson and Shiffrin (1968) multi-store model dominated memory research. This model proposed that information flows through three structurally distinct stores — sensory register, short-term store, and long-term store — with rehearsal serving as the mechanism for transferring information from short-term to long-term memory. The model predicted that the longer information is held in short-term memory through rehearsal, the more likely it is to be transferred to long-term storage.

Craik and Lockhart (1972) argued that this structural approach was inadequate on both empirical and theoretical grounds. They proposed instead that memory is a byproduct of the type of processing performed on incoming information, organized along a continuum from shallow (structural, surface-level analysis) to deep (semantic, meaning-based analysis). Rather than distinct memory stores, they envisioned a single processing system in which the depth of analysis at encoding determines the persistence and retrievability of the resulting memory trace.

The Framework

Craik and Lockhart (1972) proposed a continuum of processing depth with three broadly defined levels. Structural processing (the shallowest level) involves analyzing physical features — the visual form of a word, the typeface, or the spatial arrangement of letters. Phonological processing (an intermediate level) involves analyzing sound patterns — the pronunciation, rhyming properties, or rhythmic structure. Semantic processing (the deepest level) involves analyzing meaning — the conceptual content, categorical membership, or associative relationships of the stimulus.

The key theoretical claim was that deeper processing produces more elaborate, more distinctive, and more durable memory traces. Shallow processing creates traces that capture only surface features and decay rapidly. Deep processing creates traces that are richly connected to existing knowledge structures, providing multiple retrieval routes and resistance to forgetting.

Craik and Tulving (1975): The Definitive Experiments

The most thorough experimental validation of the LOP framework came from Craik and Tulving's (1975) landmark paper "Depth of Processing and the Retention of Words in Episodic Memory." Across 10 experiments, they systematically manipulated processing depth using an incidental learning paradigm — participants answered orienting questions about words without knowing a memory test would follow.

In the basic design, participants saw a word preceded by a question requiring structural analysis ("Is the word printed in capital letters?"), phonological analysis ("Does the word rhyme with WEIGHT?"), or semantic analysis ("Would the word fit in the sentence: 'He met a ___ in the street'?"). On a subsequent unexpected recognition test, semantic words were recognized at rates of 0.81, phonological at 0.63, and structural at 0.15 — a massive and orderly effect of processing depth (Experiment 1).

Subsequent experiments in the same paper demonstrated that: (a) the depth effect was not due to differential rehearsal time, since structural questions actually took longer to answer than semantic questions in some conditions; (b) the effect held for both recognition and recall; (c) richly elaborated semantic encodings (complex sentence frames) produced better memory than minimal semantic encodings (simple category judgments); and (d) the effect was not attributable to differential effort or difficulty, since equating processing time across conditions preserved the depth advantage.

Maintenance vs. Elaborative Rehearsal

A critical implication of the LOP framework is that not all rehearsal is equal. Craik and Watkins (1973) demonstrated this by asking participants to maintain words in short-term memory for varying durations through rote repetition (maintenance rehearsal). Despite some words being rehearsed for up to 18 seconds and others for only 2 seconds, subsequent recall was unrelated to rehearsal duration. This result directly contradicted the Atkinson-Shiffrin prediction that longer rehearsal produces better long-term storage.

The distinction between maintenance rehearsal (Type I — rote repetition that maintains information in short-term memory without strengthening long-term storage) and elaborative rehearsal (Type II — processing that connects new information to existing knowledge through meaningful associations) became central to the LOP framework. Only elaborative rehearsal, which involves deeper semantic processing, reliably improves long-term retention. A student who reads a definition 10 times (maintenance rehearsal) will remember less than one who generates a single personal example of the concept (elaborative rehearsal).

The Self-Reference Effect

Rogers et al. (1977) extended the depth-of-processing framework by demonstrating that self-referential processing ("Does this word describe you?") produces even better memory than standard semantic processing ("Does this word mean the same as ___?"). Across their experiments, self-referential encoding yielded recall rates approximately 20-30% higher than semantic encoding, suggesting that the self-concept provides a uniquely rich and well-organized knowledge structure for deep encoding.

Symons and Johnson's (1997) meta-analysis of 129 self-reference studies confirmed the robustness of this effect across diverse materials, populations, and retention intervals. The self-reference advantage over semantic processing is moderate in size (d ≈ 0.50) but remarkably consistent, and it has practical implications for education: students who relate course material to their own experiences show enhanced retention compared to those who process it only for general meaning.

The Generation Effect

Slamecka and Graf (1978) demonstrated a closely related phenomenon: words that participants generate themselves (completing "OPPOS___" when given the cue "HOT") are remembered better than words they simply read ("HOT-COLD"). This generation effect, replicated across hundreds of studies, is consistent with the LOP framework because generating a word requires deeper semantic processing — accessing meaning, evaluating conceptual relationships, and producing the target — than passively reading it.

The generation effect has been integrated into educational practice through techniques like elaborative interrogation (generating explanations for why facts are true) and the testing effect (generating answers from memory rather than restudying). Both techniques force deeper processing during learning, consistent with the core LOP principle that active, meaning-based engagement produces stronger memories than passive exposure.

Transfer-Appropriate Processing

The most significant theoretical challenge to the LOP framework came from Morris et al. (1977), who demonstrated that depth of processing does not universally predict memory performance. In their critical experiment, participants encoded words using either semantic orienting questions or rhyme orienting questions, then received either a standard recognition test or a rhyme recognition test ("Did any word in the study list rhyme with LEGAL?").

The results were striking: on the standard recognition test, semantic encoding produced superior performance, replicating the classic LOP effect. But on the rhyme recognition test, phonological encoding produced superior performance — a complete reversal. This demonstrated that memory performance depends not on depth of encoding per se but on the match between encoding processes and retrieval demands. The principle of transfer-appropriate processing states that memory is best when the cognitive processes engaged at retrieval overlap with those engaged at encoding.

Transfer-appropriate processing does not invalidate the LOP framework but qualifies it importantly. In everyday life, most retrieval situations require access to meaning (understanding a conversation, applying knowledge to a problem, answering an exam question), which is why semantic processing generally produces the best memory. The LOP framework holds as a practical heuristic precisely because most real-world retrieval is meaning-based. But the deeper principle is not "depth" per se but encoding-retrieval compatibility — as formalized in Tulving's (1983) encoding specificity principle.

Distinctiveness and Elaboration

Hunt and Einstein (1981) proposed that the LOP effect reflects two separable encoding processes: elaboration (the quantity and richness of encoding) and distinctiveness (the extent to which an encoded item is different from other items in memory). Semantic processing produces both greater elaboration (connecting the item to more knowledge) and greater distinctiveness (creating a unique encoding that stands out from competing traces).

This distinction explains anomalies that pure "depth" cannot. Bizarre imagery, for example, does not involve deeper semantic processing than common imagery, yet it can produce better memory — because bizarre images are more distinctive. Similarly, Hunt and Worthen (2006) showed that the LOP effect can be decomposed: relational processing (identifying similarities among items) supports recall by facilitating organization, while item-specific processing (identifying unique features of individual items) supports recognition by facilitating discrimination. The most effective encoding engages both processes — a principle captured by the distinctiveness-fluency framework.

Neural Basis: The Subsequent Memory Effect

Neuroimaging has provided striking confirmation of the LOP framework. Wagner et al. (1998) demonstrated that activation of the left inferior prefrontal cortex (LIPC) and the medial temporal lobe during encoding predicts whether an item will be subsequently remembered or forgotten — the "Dm effect" (difference due to memory). Words processed semantically produce greater LIPC activation than words processed structurally, and within semantic processing, words that are later remembered produce greater activation than words that are later forgotten.

Otten et al. (2002) showed that the neural subsequent memory effect is process-specific: deep (semantic) encoding produces a subsequent memory effect in left prefrontal and medial temporal regions, while shallow (phonological) encoding produces a subsequent memory effect in different regions (left inferior temporal and fusiform cortex). This neural dissociation supports the view that qualitatively different encoding processes — not just "more" or "less" of the same process — underlie the LOP effect.

Kim (2011) conducted a meta-analysis of fMRI studies of encoding depth, confirming that deeper processing consistently activates left inferior frontal gyrus (semantic retrieval and selection), left middle temporal gyrus (semantic knowledge storage), and bilateral hippocampal formation (episodic memory binding). The convergence of neuroimaging data with behavioral findings strengthens the theoretical foundations of the LOP framework.

Applications to Education

The LOP framework has had an enormous impact on educational practice. Effective study strategies are, at their core, strategies for inducing deep processing:

• Elaborative interrogation — Asking "why" and "how" questions about facts forces semantic processing and integration with prior knowledge. Dunlosky et al. (2013) rated it as having moderate utility across a wide range of learning conditions.
• Self-referential encoding — Relating material to personal experiences engages the deepest level of semantic processing through the richly organized self-schema.
• Retrieval practice — Testing oneself forces generation of answers, which engages deeper processing than rereading. Roediger and Karpicke (2006) demonstrated that retrieval practice produces superior long-term retention compared to additional study time.
• Concept mapping — Creating visual diagrams of relationships forces semantic processing by requiring learners to identify and articulate meaningful connections.
• Teaching others — Explaining material to someone else forces deep reorganization and semantic processing. The "protégé effect" (Nestojko et al., 2014) shows that even the expectation of teaching enhances encoding depth.

Conversely, the LOP framework explains why common study strategies are ineffective. Highlighting, rereading, and copying notes are all forms of shallow, structural processing that engage minimal semantic analysis. Dunlosky et al. (2013) rated highlighting and rereading as having low utility precisely because they fail to induce deep processing — students process surface features (spatial location on the page, visual appearance of highlighted text) without engaging with meaning.

Aging and Levels of Processing

Craik and colleagues have shown that the depth-of-processing effect interacts importantly with aging. Craik and Byrd (1982) proposed that aging reduces the cognitive resources available for self-initiated deep processing. Older adults show reduced spontaneous use of semantic encoding strategies but benefit normally (or even disproportionately) when deep processing is experimentally induced through orienting tasks. This "environmental support" hypothesis suggests that the elderly memory deficit is partly an encoding deficit — older adults do not spontaneously engage in the deep processing that younger adults perform automatically, but they can be prompted to do so with appropriate task demands.

Craik and Rose (2012) demonstrated that the depth-of-processing effect remains robust across the lifespan, with similar proportional benefits of deep over shallow processing in younger and older adults. This has practical implications: environmental supports that induce deep processing (structured study guides, elaborative questions, retrieval practice) can substantially reduce age-related memory decline by compensating for reduced self-initiated processing.

Criticisms and Ongoing Debates

Despite its enormous influence, the LOP framework has been subject to several important criticisms that have shaped its development:

• Circularity — Baddeley (1978) argued that "depth" is defined circularly: deep processing is whatever produces better memory, and better memory is evidence of deep processing. Without an independent measure of depth, the framework risks being unfalsifiable. Craik (2002) acknowledged this concern and suggested that processing time, neuroimaging activation patterns, and the nature of the required judgment can serve as converging measures of depth independent of memory outcomes.
• Transfer-appropriate processing — As described above, Morris et al. (1977) showed that the match between encoding and retrieval processes matters more than depth per se. This qualification limits the generality of the original framework but does not negate its practical value for the vast majority of real-world learning situations.
• Elaboration vs. depth — Craik and Tulving (1975, Experiment 8) themselves showed that richly elaborated semantic encodings ("The great bird swooped down and carried off the struggling ___") produced better memory than minimal semantic encodings ("She cooked the ___"), even though both involve semantic processing. This suggested that elaboration within a given level of depth matters, complicating the simple depth continuum.
• Intentional vs. incidental learning — The LOP effect is strongest under incidental learning conditions. When participants know a memory test is coming (intentional learning), they can adopt deep processing strategies spontaneously, potentially washing out differences between orienting conditions. This suggests that the LOP framework primarily captures encoding processes that operate automatically or by default rather than strategic, controlled memorization.

Legacy and Influence

The LOP framework, despite its limitations, remains one of the most productive ideas in the history of memory research. Craik (2002) estimated that the original 1972 paper had been cited over 4,000 times (now exceeding 10,000 citations). Its most enduring contribution was shifting the field's focus from structural questions (how many stores?) to processing questions (what operations produce good memory?). This shift opened up productive research programs on elaboration, distinctiveness, transfer-appropriate processing, the generation effect, and encoding-retrieval interactions that continue to yield insights. The practical message — that meaningful, active engagement with material produces far better learning than passive repetition — has become a cornerstone of evidence-based educational practice.

Sensory Modalities and Depth of Processing

The depth-of-processing framework applies across sensory modalities, though the nature and strength of the effect varies by modality. Different sensory channels afford different opportunities for deep encoding, and the interaction between modality and processing depth has important implications for how information is best presented for learning.

Vision

Visual input produces the strongest recall of all sensory modalities and permits the widest range of depth-of-processing manipulations. Within visual encoding, pictures consistently yield better memory than words — the picture superiority effect — because pictures automatically engage both structural and semantic processing. The richness of visual stimuli allows encoding at multiple levels simultaneously: physical features (color, shape, spatial layout), categorical identity, and semantic meaning.

Audition

Auditory stimuli follow conventional levels-of-processing rules, although general recall for auditory input is somewhat weaker than for visual input. Within auditory encoding, semantic analysis (processing the meaning of spoken words or the emotional content of music) produces the strongest retention, consistent with the depth principle. Phonological processing — attending to sound patterns, rhyme, and prosody — occupies an intermediate level.

Touch

Tactile memory representations arising from the somatosensory system are similar in structure to visual representations, with evidence for both shallow (texture, temperature) and deep (object identity, functional meaning) processing levels. Haptic exploration of objects can produce robust memory traces, particularly when touch is combined with visual encoding to create multimodal representations.

Smell

Olfactory memory is weaker than visual memory, achieving successful identification rates of roughly 70–80% of those for visual stimuli. However, levels-of-processing effects emerge within olfactory encoding when participants are asked to visualize the source of the odor or to generate a semantic label — for example, attributing a muddy odor to "smells like a rain puddle." Semantic processing of odors (naming, categorizing, associating with autobiographical memories) significantly improves subsequent recognition, demonstrating that the depth principle extends even to the phylogenetically oldest sensory system.

Key Researchers

• Fergus I. M. Craik — Co-creator of the levels-of-processing framework (1972) and central figure in its subsequent development, including work on maintenance vs. elaborative rehearsal, encoding-retrieval interactions, and aging effects on depth of processing.
• Robert S. Lockhart — Co-author of the original 1972 LOP framework paper with Craik, proposing that memory is a byproduct of processing depth rather than structural storage location.
• Endel Tulving — Co-authored the definitive 1975 experimental validation of the LOP framework with Craik and later formalized the encoding specificity principle, which extended depth-of-processing ideas into a broader theory of encoding-retrieval compatibility.
• Richard C. Atkinson & Richard M. Shiffrin — Developed the multi-store model of memory (1968) that the LOP framework was explicitly designed to challenge and supplant.
• Michael J. Watkins — Demonstrated with Craik (1973) that maintenance rehearsal does not improve long-term retention, providing critical evidence against the multi-store model's rehearsal-transfer assumption.
• Timothy B. Rogers, Nicholas A. Kuiper & W. S. Kirker — Discovered the self-reference effect (1977), showing that relating information to the self produces even deeper encoding than standard semantic processing.
• Cynthia S. Symons & Blair T. Johnson — Conducted the definitive meta-analysis (1997) confirming the robustness of the self-reference effect across 129 studies.
• Norman J. Slamecka & Peter Graf — Demonstrated the generation effect (1978), showing that self-generated information is remembered better than passively read information — consistent with deeper processing during generation.
• C. Donald Morris, John D. Bransford & Jeffery J. Franks — Proposed transfer-appropriate processing (1977), the most significant theoretical challenge to the LOP framework, showing that encoding-retrieval match matters more than depth per se.
• R. Reed Hunt & Gilles O. Einstein — Proposed (1981) that the LOP effect reflects two separable processes — elaboration and distinctiveness — refining the theoretical basis beyond a simple depth continuum.
• Anthony D. Wagner & Daniel L. Schacter — Demonstrated the neural subsequent memory effect (1998), showing that left prefrontal and medial temporal activation during encoding predicts later remembering.
• Leun J. Otten, Richard N. A. Henson & Michael D. Rugg — Showed (2002) that the neural subsequent memory effect is process-specific, with deep and shallow encoding producing memory-predictive activity in different brain regions.
• Hongkeun Kim — Conducted a meta-analysis (2011) of 74 fMRI encoding studies, confirming the neural regions consistently associated with depth-of-processing effects.
• John Dunlosky — Led a comprehensive review (2013) of learning techniques that translated LOP principles into evidence-based educational recommendations.
• Henry L. Roediger III & Jeffrey D. Karpicke — Demonstrated (2006) that retrieval practice produces superior long-term retention, connecting the testing effect to deeper processing during generation.
• Alan D. Baddeley — Raised the influential circularity criticism (1978), arguing that "depth" lacked an independent definition — a challenge that shaped the framework's subsequent theoretical refinement.
• Nathan S. Rose — Collaborated with Craik (2012) to demonstrate that depth-of-processing effects remain robust across the lifespan, supporting environmental-support interventions for age-related memory decline.
• Laird S. Cermak & Loraine Reale — Investigated (1978) depth-of-processing effects in amnesic patients, revealing how hippocampal damage disrupts the binding of deeply processed information into episodic memory.

Disorders

• Depression — Reduced spontaneous engagement in deep, elaborative processing; depressed individuals show a negativity bias in semantic encoding, preferentially processing negative self-referential material (Mathews & MacLeod, 2005)
• Amnesia — Hippocampal damage impairs the binding of deeply processed features into coherent episodic traces, though the depth-of-processing manipulation can still produce graded effects in mild amnesia (Cermak & Reale, 1978)
• Alzheimer's disease — Progressive semantic memory degradation undermines the knowledge base needed for deep processing; early-stage patients show reduced LOP effects as semantic processing becomes impoverished
• ADHD — Attentional deficits reduce the likelihood of sustained, deep processing during encoding; individuals with ADHD benefit normally from depth manipulations when processing is experimentally induced but fail to self-initiate deep strategies
• Schizophrenia — Disorganized semantic networks impair the quality of deep processing, producing reduced LOP effects particularly for self-referential encoding

References