Cognitive Connie
Understanding craik & tulving — levels of processing
Fergus Craik and Endel Tulving's 1975 paper presented one of the most influential challenges to the multi-store model of memory. Rather than asking how memory is structured (boxes and arrows of stores), they asked how memory is created — what happens at the moment of encoding that determines how well something will be remembered.
Defining features
Key figures
Endel Tulving
1927–2023Estonian-Canadian cognitive neuroscientist and one of the most influential memory researchers of the 20th century. His distinction between episodic and semantic memory, encoding specificity principle, and synaptic theory of memory consolidation built on and interacted productively with the levels-of-processing approach co-developed with Craik.
Fergus I. M. Craik
1935–British-Canadian cognitive psychologist who (with Robert Lockhart, 1972) proposed the levels-of-processing framework and (with Tulving, 1975) provided its most influential empirical demonstration. Has continued to develop and refine the framework across five decades of memory research, including work on aging and working memory.
Methods
The orienting task paradigm
Craik and Tulving devised an incidental learning paradigm — participants memorised nothing intentionally, yet produced systematic memory differences.
Three levels of orienting questions
Each word was preceded by a question requiring structural processing (Is this word in upper case?), phonemic processing (Does this rhyme with X?), or semantic processing (Does this fit the sentence 'He met a ___ in the street'?). Questions were answered yes or no, and the word was then shown for ~200 ms. Participants performed 60 such trials believing the study was about reaction time or perceptual processing.
Incidental learning design
Participants were not told to memorise the words. Memory was tested only after all orienting questions were complete — by surprise cued recall or recognition. This design separates encoding strategy from intentional memorisation, showing that memory quality depends on what you do with material, not whether you intend to remember it.
Yes/no compatibility manipulation
Across 10 experiments, Craik and Tulving also varied whether the correct answer to an orienting question was yes or no (e.g., a word that fits a sentence vs. one that does not). Words that produced a yes response (i.e., fit the semantic frame) were consistently better remembered than no-answer words at the same depth — suggesting that elaboration and congruence with existing knowledge further boosted encoding.
Findings
What the experiments showed
Levels effect: semantic > phonemic > structural
Recognition memory after semantic orienting was approximately 65–70%, after phonemic ~35–40%, and after structural ~15–20% — a roughly fourfold superiority of semantic over structural processing. This gradient was replicated across all 10 experiments in the paper using different materials, question types, and retention tests.
Intent to learn is not the key variable
An intentional learning control group (told to remember the words, no orienting questions) performed no better than the semantic-orienting incidental group. What mattered was the depth and elaborateness of processing, not conscious memorisation effort. This was a direct challenge to the idea that transferring material to LTM is a deliberate act.
Elaboration within semantic level
Yes answers (word fits the semantic frame) produced better memory than no answers even when both required semantic processing. Craik and Tulving argued that elaboration — how richly the word is integrated into a semantic context — adds a further dimension beyond raw processing depth.
Criticism
Limitations and debates
Circular definition of 'depth'
The original levels-of-processing framework cannot independently measure processing depth — depth is inferred from memory performance, but memory performance is what the framework is supposed to explain. This circularity means the theory cannot be falsified in the strong sense: any memory difference can be attributed post hoc to differences in depth.
Transfer-appropriate processing (Morris et al., 1977)
Morris, Bransford, and Franks showed that "shallow" rhyme-based encoding actually produced better memory on a rhyme recognition test than semantic encoding did. This transfer-appropriate processing finding showed that what matters is not absolute depth but the match between encoding and retrieval conditions — a significant revision of the original claims.
Time confounded with depth
Semantic orienting questions take longer to answer than structural ones. It is possible that slower, more time-consuming processing simply gives more time for encoding, and that time — not semantic content — is the critical variable. Later studies tried to equate response times across levels, with mixed success.
Relevance for today
Why it still matters
Transfer-appropriate processing
Morris, Bransford, and Franks (1977) extended the paradigm by showing that "shallow" phonemic encoding can outperform "deep" semantic encoding when the retrieval test is phonemic — demonstrating that what matters is the match between encoding and retrieval context, not depth per se. This transfer-appropriate processing principle reconciled levels of processing with Tulving's encoding specificity account and remains a cornerstone of contemporary memory theory.
Self-referential processing effect
Rogers, Kuiper, and Kirker (1977) extended the paradigm by adding a self-referential level ("Does this word describe you?"): self-referential encoding produced the best recall of all. This self-reference effect is now a well-replicated finding with implications for clinical psychology — people with depression show a diminished self-reference memory advantage.
Reconceptualised as encoding specificity
The later transfer-appropriate processing framework repositioned levels of processing within Tulving's encoding specificity principle: memory depends on the overlap between encoding and retrieval contexts. This synthesis between Craik's framework and Tulving's encoding specificity work has become the dominant framework for understanding why some learning strategies work better than others.
Test your knowledge
Frequently asked questions
What is the levels of processing effect?+
The finding that the depth at which we process information at encoding determines how well we remember it. Semantic processing — engaging with the meaning of material — produces stronger, more durable memories than phonemic (sound-based) or structural (surface-feature) processing, even when the intent to memorise is identical or absent.
Why is the levels of processing framework criticised?+
The main criticism is circularity: depth is defined by the quality of memory it produces, and memory quality is explained by depth. There is no independent measure of depth, making the theory difficult to falsify. Transfer-appropriate processing (Morris et al., 1977) also showed that "shallow" encoding can produce better memory than "deep" encoding when the retrieval test matches the encoding modality.
Sources
Last reviewed August 2026- 1.
Craik, F. I. M., & Tulving, E. (1975). Depth of processing and the retention of words in episodic memory. Journal of Experimental Psychology: General, 104(3), 268–294. https://doi.org/10.1037/0096-3445.104.3.268
+About this source
The original 10-experiment paper demonstrating the levels-of-processing effect across structural, phonemic, and semantic orienting tasks with incidental learning design.
- 2.
Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11(6), 671–684. https://doi.org/10.1016/S0022-5371(72)80001-X
+About this source
Theoretical paper proposing the levels-of-processing framework; the conceptual precursor to the 1975 empirical study.
- 3.
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
+About this source
Demonstrates transfer-appropriate processing — the key revision to the levels framework showing that encoding-retrieval match matters more than absolute depth.