The levels of processing (LOP) model, proposed by Craik and Lockhart (1972), holds that how well something is remembered depends on the depth of processing applied to it at encoding.
It does not depend on which memory store it enters or how long it is rehearsed. Shallow, surface-level analysis leaves a weak, short-lived memory trace, while deep, meaning-based analysis leaves a durable one. Depth of processing is what predicts memory.
Craik and Lockhart (1972) suggested a continuum of processing depth, ranging from shallow sensory analysis to deep semantic analysis. This continuum is generally divided into three distinct levels:
Key Takeaways
- Depth of Processing: Memory is a by-product of how deeply information is analysed at encoding, not which store it passes through or how long it is rehearsed (Craik & Lockhart, 1972).
- Three Levels: Processing runs from shallow (structural/visual) through intermediate (phonemic/sound) to deep (semantic/meaning); deeper processing produces longer-lasting memories.
- Elaboration & Distinctiveness: It isn’t depth alone: how richly elaborated (Craik & Tulving, 1975) and how distinctive (Bransford et al., 1979) an encoding is also shapes how well it’s remembered.
- Self-Reference Effect: Material related to yourself is remembered best of all, because the self is an unusually rich, distinctive knowledge structure (Rogers et al., 1977).
- Key Criticism: The theory has been accused of circularity: “depth” is inferred from good recall, then used to explain that same recall (Baddeley, 1978).
- Modern Evidence: Recent meta-analyses suggest “depth” works because it taps adaptive, survival-relevant processing, not depth for its own sake (Scofield et al., 2018).
| Level | Type of analysis | Example orienting question | Depth |
|---|---|---|---|
| Structural (shallow) | Surface/physical features | “Is the word in CAPITAL letters?” | Shallowest |
| Phonemic (intermediate) | The word’s sound | “Does the word rhyme with weight?” | Intermediate |
| Semantic (deep) | The word’s meaning | “Would it fit the sentence ‘He met a ___ in the street’?” | Deepest |
Shallow Processing
Shallow processing involves superficial or surface-level encoding of information. It typically focuses on sensory features or basic characteristics without engaging in meaningful analysis or elaboration.
Shallow processing only involves maintenance rehearsal (repetition to help us hold something in the STM) and leads to fairly short-term retention of information.
As a result, shallow processing leads to poorer memory encoding and weaker retention than deep processing, which involves more thorough and meaningful engagement with the information.
Structural processing (appearance): which is when we encode only the physical qualities of something. E.g. the typeface of a word or how the letters look. For example, when looking at a word, a person might only pay attention to the shapes of the letters, count the number of vowels, or determine whether the word is written in uppercase or lowercase letters
Intermediate Phonemic Processing
Intermediate phonemic processing (also referred to as phonological processing) represents the middle tier of cognitive depth in the Levels of Processing (LOP) framework proposed by Craik and Lockhart.
At this level, analysis moves beyond a stimulus’s physical or visual characteristics (shallow structural processing). It begins to translate physical shapes into meaningful auditory units, attaching phonetic sounds to them.
Phonemic processing: processing here focuses on how a word sounds, such as identifying whether a target word rhymes with another word.
Intermediate phonemic processing is essential to short-term memory capacity.
In Baddeley and Hitch’s model of working memory, this is handled by the phonological loop. The loop consists of a passive phonological store and an active articulatory rehearsal process: the “inner voice” that repeats information.
Deep Processing
Deep processing refers to the meaningful and thorough encoding of information.
It involves engaging with the content thoughtfully and elaborately, making connections to existing knowledge and personal experiences.
Deep processing promotes better memory retention and recall than shallow, surface-level processing.
Craik put this precisely.
Craik (1973, p. 48) defined depth not as the number of operations performed on a stimulus, but as “the meaningfulness extracted from the stimulus.” That distinction shifts the focus from quantity of processing to quality of engagement.
Meaning changes everything.
The primary mechanism is semantic processing: relating incoming information to previously stored meanings or personal experiences.
When a learner considers a word’s implications rather than merely its sound or appearance, they produce a more durable memory trace. That trace resists decay because it sits within a dense network of existing associations, rather than in isolation.
elaborative rehearsal:
Deep processing depends on elaborative rehearsal: the active transformation of material to make it more meaningful.
This goes beyond simple repetition.
The learner expands on the information by constructing mental images, identifying logical connections, or linking new content to prior knowledge.
Each layer of meaning added strengthens the memory trace and integrates the new material into a coherent knowledge structure, making subsequent retrieval more reliable.
Deep semantic encoding leads to richer memory codes. That’s largely because elaboration is easier to achieve at the semantic level than at shallower structural or phonological levels.
The Self-Reference Effect
The deepest, most elaborative form of semantic encoding is to relate material to yourself.
Rogers, Kuiper, and Kirker (1977) had participants judge trait adjectives under four orienting tasks: structural, phonemic, semantic, and self-referent. The self-referent task simply asked: does this word describe you?
Words encoded with reference to the self were recalled best of all: the self-reference effect (SRE). That’s because the self is an exceptionally rich, distinctive knowledge structure to bind information to. It combines maximal elaboration with maximal distinctiveness.
A meta-analysis of 129 studies confirmed a reliable memory advantage for self-referent over semantic encoding (Symons & Johnson, 1997). It traced the advantage to the elaborative and organisational properties of the self-concept.
Key Study: Craik and Tulving (1975)
Aim
To investigate whether the depth of processing affects the long-term recognition of words.
A second strand tested whether recognition also improves with how elaborate that semantic processing is, not just how deep it is.
Method
Participants were presented with 60 words and asked questions requiring structural, phonological, or semantic analysis. A surprise recognition test followed these tasks.
Some questions required the participants to process the word in a deep way (e.g. semantic) and others in a shallow way (e.g. structural and phonemic). For example:
-
Shallow graphemic (structural) processing: Participants were asked to focus on the physical and visual characteristics of the words, such as deciding whether a given word was printed in uppercase or lowercase letters.
-
Intermediate phonemic processing: Participants were asked to focus on the auditory qualities of the words, such as determining whether a presented word rhymed with a specific target word.
-
Deep semantic processing: Participants were asked to focus entirely on the meaning of the words by deciding whether each word logically fitted into a blank space within a provided sentence.
Participants were then given a long list of 180 words into which the original words had been mixed.
They were asked to pick out the original words.
elaboration experiment
Craik and Tulving (1975) also sought to explore the impact of elaboration of processing, which refers to the sheer amount or richness of cognitive processing occurring at a specific level.
To test this, they manipulated the complexity of the sentences used in the deep semantic task.
Participants were shown a word and asked whether it fitted a sentence frame. One version was simple: “She cooked the ____“. The other was complex and descriptive: “The great bird swooped down and carried off the struggling ____“.
Results
Recognition rose steeply with depth: roughly 16% of structurally processed words, 57% of phonemically processed words, and 83% of semantically processed words were later recognised.
The elaboration experiment showed a similar pattern. Cued recall was twice as high for words that had accompanied the complex sentences as for those in the simple sentences.
Deeply processing a word’s meaning was effective on its own. Embedding that meaning in a richer context made the resulting memory trace stronger still.
There was a secondary finding too. Words that had produced a “Yes” response were remembered better than those producing a “No”, suggesting a coherent, integrated encoding matters as well as depth.
This same pattern held whether the researchers measured recognition or free recall afterwards.
Conclusion
Semantically processed words involve elaborative rehearsal and deep processing, which produce more accurate recall than the shallow processing behind phonemic and visual encoding.
The incidental design is the study’s great strength. Because participants did not know they would be tested, the depth effect cannot be explained away as people simply choosing to rehearse the “meaning” words more.
The design also quietly undermines the multi-store model’s one-way flow. Judging whether a word fits a category means consulting knowledge already held in long-term memory, so encoding depends on long-term memory too, not just the other way round.
Its main limit is that depth and effort are hard to separate, since semantic questions usually take longer to answer. This confound is explored further below.
A Second Key Study: Hyde and Jenkins (1973)
Aim: to test whether it is meaning-based (semantic) processing, rather than the intention to learn, that determines how much of a word list is recalled.
Method: participants performed an orienting task on a word list without knowing a memory test would follow: either semantic (rating pleasantness) or non-semantic (counting letters). A surprise free-recall test followed.
Findings: words processed by a semantic orienting task were recalled far better than words processed non-semantically. Semantic incidental learners even recalled as many words as those who had intended to learn all along. Intending to learn barely mattered.
Conclusion: what matters is the nature of the processing, not the intention to memorise. Meaning-based processing under incidental conditions alone is enough to produce excellent recall.
This design cleanly separates intention from depth. That makes it some of the strongest support for the model. Its limitation is shared with Craik and Tulving’s study: the tasks may simply be more distinctive or effortful than the surface ones.
Real-Life Applications
This explanation of memory is useful in everyday life because it highlights the way in which elaboration, which requires deeper processing of information, can aid memory.
Three examples of this are.
- Reworking – putting information in your own words or talking about it with someone else.
- Method of loci – when trying to remember a list of items, linking each with a familiar place or route.
- Imagery – by creating an image of something you want to remember, you elaborate on it and encode it visually (i.e. a mind map).
All three examples can be used to revise psychology using semantic processing (e.g. explaining memory models to your mum, or using mind maps). Using elaboration rehearsal like this should result in deeper processing.
Consequently, more information will be remembered (and recalled) and better exam results should be achieved.
Critical Evaluation
Strengths
A central strength of the LOP approach is its assumption that perception, attention, and memory are closely interconnected.
Rather than treating memory as an isolated system of rigid storage boxes, the theory correctly posits that learning and remembering are natural by-products of perception, attention, and comprehension.
This holistic view accurately reflects the fluid and integrated nature of human cognition.
The theory successfully identified elaboration and distinctiveness as crucial determinants of memory.
The LOP framework established that long-term retention depends heavily on the depth of analysis. It also highlighted that enriching information (elaboration) and making it unique (distinctiveness) are critical for strong memory formation.
Functional neuroimaging has provided neurobiological backing for the theory’s claims about depth.
Functional neuroimaging provides biological evidence that deep processing engages specific regions of the human brain.
Researchers utilize Functional Magnetic Resonance Imaging (fMRI): a technology that measures brain activity by detecting changes associated with blood flow.
These studies reveal that semantic tasks trigger higher metabolic activity in the frontal and temporal lobes compared to shallow tasks.
Wagner et al. (1998)
-
Aim: To identify the neural correlates associated with the depth of processing during encoding.
-
Procedure: Participants were scanned using fMRI while they performed either semantic or perceptual tasks on various words.
-
Findings: Increased activation was observed in the left inferior frontal lobe and the medial temporal lobes during semantic processing.
-
Conclusions: Semantic encoding recruits specialized neural circuits that facilitate the formation of robust and accessible memory traces.
Weaknesses
Lack of explanatory depth
Eysenck (1990) argued that the theory describes rather than explains.
Craik and Lockhart demonstrated that deep processing produces better long-term memory than shallow processing, but offered no detailed account of why this is the case.
Subsequent research has partly addressed this: deeper coding seems to help memory because it is more elaborate, activating more semantic associations and weaving new material into existing knowledge.
That only partly closes the gap.
Craik (2002) himself later conceded that “depth” was always more a useful heuristic than a precise mechanism.
The underlying mechanism remains only partly specified.
The vagueness of “depth”
The theory’s central construct cannot be independently observed or objectively measured.
One critic pressed this hardest. Baddeley (1978) argued that the “levels” metaphor could not be operationalised precisely enough to work as a real explanatory theory.
Without a way to measure processing depth separately from memory itself, the framework risks circularity. Deep processing is inferred from better recall, which is then explained by deep processing.
That critique has never been fully answered.
Effort as a confound
Deeper processing typically demands more cognitive effort than shallow processing.
It is therefore unclear whether superior retention reflects the meaningfulness of the encoding or simply the effort invested.
Depth, time, effort, elaboration, and distinctiveness are hard to prise apart. That is exactly why researchers increasingly talk about elaboration and distinctiveness rather than “depth” itself.
The role of distinctiveness
Research by Bransford et al. (1979) revealed that depth and elaboration are not the only determinants of retention.
Distinctiveness matters too. The sentence “A mosquito is like a doctor because both draw blood” is better recalled than the more elaborated version.
That’s a distinctive image.
“A mosquito is like a racoon because they both have heads, legs and jaws” is that more elaborated alternative. The first sentence is less elaborate but more distinctive: its incongruity makes it memorable.
This finding suggests that the theory underestimates the independent contribution of distinctiveness to memory formation.
Neglect of retrieval
The assumption that deep processing is always superior was later challenged.
Aim: to test whether semantic processing always wins. Or whether its advantage instead depends on the type of memory test used at retrieval.
Method: participants answered either semantic questions (about a word’s meaning) or phonemic questions (whether it rhymed with another word). Memory was tested two ways. One was a standard recognition test; the other was a rhyming test that rewarded phonemic information instead.
Results: on the standard test, semantic processing produced better recognition, replicating the usual finding. Then the pattern reversed. Words processed phonemically were recognised better than words processed semantically on the rhyming test.
Conclusion: there is no fixed hierarchy in which deep always beats shallow. What matters is whether the processing done at encoding suits the processing the retrieval test demands.
This is decisive counter-evidence. Morris et al. (1977) undercuts any strong “depth-always-wins” reading of the theory.
It does not abolish the depth effect: deep encoding is still best for meaning-based tests, which dominate everyday life. It simply shows the effect is a special case of encoding-retrieval match, not a universal law.
Implicit memory and amnesia
Levels-of-processing effects are considerably stronger in explicit memory than in implicit memory.
More problematically, the theory struggles to account for amnesic patients. They retain intact semantic processing abilities yet show severely impaired long-term memory, a dissociation the depth-of-processing account has no ready explanation for.
Theoretical Revision
In response to these criticisms, Lockhart and Craik (1990) revised their original framework.
They acknowledged that the encoding-retrieval relationship had been inadequately theorised, and conceded that deeper processing is not universally superior across all tasks and test conditions. They also accepted that the original model was overly simplistic.
The core principle nonetheless retains its standing as a foundational heuristic in cognitive psychology:
Meaningful engagement with material — actively constructing understanding rather than passively rehearsing — remains one of the most reliable routes to durable memory.
Contemporary Research
Modern research has mostly stopped asking whether deep processing helps memory.
That question is settled; the newer question is what “depth” really is, and why meaning-based encoding works so well.
The Survival-Processing Effect
The strongest recent evidence reframes deep processing as survival-relevant processing.
Nairne, Thompson, and Pandeirada (2007) found that rating words for their relevance to a survival scenario produced better retention than several other “deep” encodings. That included rating a word’s pleasantness and relating it to oneself.
A 2018 meta-analysis put that effect to a stricter test. Scofield, Buchanan, and Kostic combined traditional meta-analytic methods with several bias-correction techniques across the accumulated survival-processing literature.
The corrected effect was real but smaller than earlier estimates suggested: medium to large in size, with survival processing still routinely out-performing established “deep” encodings such as self-reference.
A similar pattern holds for animacy.
A three-level meta-analysis found that words for living things are remembered better than words for non-living things (Cheng et al., 2025). Like survival processing, it points to what mattered for survival, not simply how “deep” processing is.
Encoding-Retrieval Match and the Brain
The transfer-appropriate processing finding remains the benchmark test of the depth idea. It showed that encoding and retrieval must match, not simply run “deep” (see Transfer-appropriate processing above). Most contemporary accounts now treat the classic levels effect as a special case of that match, not a general law of memory.
Brain imaging adds a different kind of evidence.
Since Wagner and colleagues’ (1998) brain-imaging work, deep semantic encoding has been tied to a distinct, stronger neural signature in the left prefrontal and medial-temporal regions than shallow encoding. This grounds the behavioural depth effect in identifiable brain activity, without fully explaining why deep processing works.
Across this research, one pattern holds.
The basic finding that meaning-based processing aids memory survives, but “depth” itself is being replaced by more precise ideas: elaboration, distinctiveness, encoding-retrieval match, and adaptive relevance.
How does LOP theory differ from multi-store models?
While multi-store models attempt to map out the rigid architecture of where memories live, the LOP theory focuses entirely on the cognitive activities that create memories in the first place.
Structure vs. Process
The most significant difference between the two approaches is their primary focus. Multi-store models, such as the highly influential model proposed by Atkinson and Shiffrin (1968), emphasise the structural architecture of memory.
Structure comes first for them.
They propose that memory consists of fixed, permanent structural components—specifically, sensory stores, a short-term store (STS), and a long-term store (LTS). In this view, memory is heavily compartmentalised, and the primary goal of the theory is to distinguish between these different storage systems.
LOP rejects that starting point.
The Levels of Processing theory, proposed by Craik and Lockhart (1972), instead focuses on the mental processes that occur during learning. It heavily criticises multi-store models for emphasising structure at the expense of processing.
Process comes first instead.
LOP theorists argue that memory is simply a by-product of perception, attention, and comprehension. According to Craik and Lockhart, the multi-store model has the relationship between structure and process “essentially the wrong way round”.
That reversal is the whole point.
Instead of focusing on where a memory is stored, LOP theory asserts that the structural components of memory are merely the resulting consequences of perceptual analysis and information processing.
While LOP theorists do not explicitly deny the existence of different memory stores, they largely ignore them to focus on how information is encoded early on.
Mechanisms of Transfer: Rehearsal vs. Depth
The two theories also diverge sharply on the mechanisms required to retain information long-term.
The Multi-Store Model’s Reliance on Maintenance Rehearsal:
In the multi-store model, the short-term store acts as a necessary gateway or relay station. Information flows in a unidirectional, linear order from sensory memory to the STS, and then to the LTS.
Rehearsal is the gatekeeper.
The key control process that dictates whether information makes it from the short-term to the long-term store is rote rehearsal.
The multi-store model assumes a direct relationship between the amount of rehearsal an item receives in the STS and the strength of the resulting memory trace in the LTS.
The LOP Theory’s Focus on Depth of Processing:
Craik and Lockhart strongly disputed the assumption that simply repeating information (maintenance rehearsal) always improves long-term memory. They argued that maintenance rehearsal does not effectively enhance long-term retention; instead, it is the kind of rehearsal or processing that matters.
The kind of rehearsal is what matters.
The central claim of LOP theory is that the depth to which information is mentally analysed during initial exposure is what determines its memorability. Depth is what counts.
Processing occurs on a continuum, ranging from shallow, sensory/structural analysis (e.g., looking at a word’s physical shape) to intermediate phonological analysis (e.g., how it sounds). At the deep end sits semantic analysis: understanding what the word means.
The deeper and more elaborative the processing during encoding, the stronger and longer-lasting the memory trace will be.
Key Study: Craik and Watkins (1973)
Aim: to test whether long-term retention depends on how long an item is held in short-term memory, or on the kind of rehearsal it receives.
Method: participants monitored word lists for “critical” words beginning with a specified letter, holding each one only until the next appeared. The delay between critical words, and hence the amount of maintenance rehearsal each received, was varied.
Rehearsal time turned out not to matter.
Results: long-term recall was unrelated to how long a word had been maintained or how many times it had been rehearsed.
Conclusion: it is the kind of rehearsal, not the amount, that determines long-term retention, directly contradicting the multi-store model’s assumption that more rehearsal means more transfer to long-term memory.
Rigidity vs. Flexibility
Finally, multi-store models are often criticised for being overly simplistic and rigid. They assume that information flows through the mind in a strict sequence, with short-term and long-term stores operating in a uniform way.
Real memory is messier than that.
For example, Atkinson and Shiffrin’s model implies that only consciously processed information in the short-term store can reach long-term memory, which struggles to explain phenomena like implicit learning.
The LOP framework is more fluid, focusing on a continuum of processing rather than strict boundaries.
It fits messier reality better.
It acknowledges that human memory is deeply intertwined with how we interact with the world. LOP suggests that long-term memory is driven by how actively we understand, elaborate on, and find distinctiveness in new material.
That matters more than just how long we hold it in a temporary mental waiting room.
References
Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and motivation (Vol. 2, pp. 89–195). Academic Press. https://doi.org/10.1016/S0079-7421(08)60422-3
Baddeley, A. D. (1978). The trouble with levels: A reexamination of Craik and Lockhart’s framework for memory research. Psychological Review, 85(3), 139–152. https://doi.org/10.1037/0033-295X.85.3.139
Bransford, J. D., Franks, J. J., Morris, C.D., & Stein, B.S.(1979). Some general constraints on learning and memory research. In L.S. Cermak & F.I.M. Craik(Eds.), Levels of processing in human memory (pp.331–354). Hillsdale, NJ: Lawrence Erlbaum AssociatesInc.
Cheng, S., Zhao, X., Yang, X., & Liu, Y. (2025). Animacy effect in the mnemonic advantage: A three-level meta-analysis. Psychonomic Bulletin & Review, 32(6), 2807–2823. https://doi.org/10.3758/s13423-025-02737-y
Craik, F. I. M. (2002). Levels of processing: Past, present… and future? Memory, 10(5–6), 305–318. https://doi.org/10.1080/09658210244000135
Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal behavior, 11, 671-684.
Craik, F.I.M., & Tulving, E. (1975). Depth of processing and the retention of words in episodic memory. Journal of Experimental Psychology: General, 104, 268-294.
Craik, F. I. M., & Watkins, M. J. (1973). The role of rehearsal in short-term memory. Journal of Verbal Learning and Verbal Behavior, 12(6), 599–607. https://doi.org/10.1016/S0022-5371(73)80039-8
Eysenck, M. W. & Keane, M. T. (1990). Cognitive psychology: a student’s handbook, Lawrence Erlbaum Associates Ltd., Hove, UK.
Hyde, T. S., & Jenkins, J. J. (1973). Recall for words as a function of semantic, graphic, and syntactic orienting tasks. Journal of Verbal Learning and Verbal Behavior, 12(5), 471–480. https://doi.org/10.1016/S0022-5371(73)80027-1
Morris, C. D., Bransford, J. D., & Franks, J. J. (1977). Levels of processing versus transfer appropriate processing. Journal of Verbal Learning and Verbal Behavior, 16(5), 519–533. https://doi.org/10.1016/S0022-5371(77)80016-9
Nairne, J. S., Thompson, S. R., & Pandeirada, J. N. S. (2007). Adaptive memory: Survival processing enhances retention. Journal of Experimental Psychology: Learning, Memory, and Cognition, 33(2), 263–273. https://doi.org/10.1037/0278-7393.33.2.263
Rogers, T. B., Kuiper, N. A., & Kirker, W. S. (1977). Self-reference and the encoding of personal information. Journal of Personality and Social Psychology, 35(9), 677–688. https://doi.org/10.1037/0022-3514.35.9.677
Scofield, J. E., Buchanan, E. M., & Kostic, B. (2018). A meta-analysis of the survival-processing advantage in memory. Psychonomic Bulletin & Review, 25(3), 997–1012. https://doi.org/10.3758/s13423-017-1346-0
Symons, C. S., & Johnson, B. T. (1997). The self-reference effect in memory: A meta-analysis. Psychological Bulletin, 121(3), 371–394. https://doi.org/10.1037/0033-2909.121.3.371
Wagner, A. D., Schacter, D. L., Rotte, M., Koutstaal, W., Maril, A., Dale, A. M., Rosen, B. R., & Buckner, R. L. (1998). Building memories: Remembering and forgetting of verbal experiences as predicted by brain activity. Science, 281(5380), 1188–1191. https://doi.org/10.1126/science.281.5380.1188
Further Information
What is the main idea of levels of processing theory?
The main idea of the levels of processing theory is that the depth at which information is processed during encoding affects its subsequent recall. According to this theory, information processed at a deeper level, such as through semantic or meaningful processing, is more likely to be remembered than information processed at a shallow level, such as through superficial or sensory-based processing.
What is deep processing?
Deep processing refers to the meaningful and thorough encoding of information. It involves engaging with the content thoughtfully and elaborately, making connections to existing knowledge and personal experiences. Deep processing promotes better memory retention and recall than shallow, surface-level processing.
