The Craik-Lockhart Levels of Processing
Read a word and decide whether it is printed in capital letters. Read another word and decide whether it fits the sentence "He met a ___ in the street." You spend about the same second or so on each. A few minutes later, without warning, someone asks you which words you saw. You will almost certainly remember the second word far better than the first.
That simple result sits at the heart of one of the most cited ideas in memory research: the levels of processing framework. Fergus Craik and Robert Lockhart, both at the University of Toronto, published it in 1972 in the Journal of Verbal Learning and Verbal Behavior. Their argument was that how well you remember something depends mainly on what kind of mental work you do with it when you first meet it. Analyzing a word's appearance leaves a weak trace; analyzing its meaning leaves a strong one.
An important correction to how this story is often told: the 1972 paper was a theoretical article, not an experiment. It reviewed existing studies and proposed a framework. The famous case-rhyme-meaning experiments came three years later, from Craik and Endel Tulving. Both papers belong to the same story, and this article covers each, along with the criticisms that followed and what the idea means for studying today.
The model it challenged: separate memory stores
In the late 1960s, the dominant account of memory was the multi-store or "modal" model, set out most influentially by Richard Atkinson and Richard Shiffrin in 1968. Information passes from brief sensory registers into a limited short-term store and, through rehearsal, into a long-term store. In its simplest popular form, the model suggested that the longer an item stays in short-term memory and the more it is rehearsed, the more likely it is to be transferred to long-term memory.
Craik and Lockhart argued that this structural picture had problems. Different studies gave very different estimates of short-term memory's capacity and duration, depending on the material and task. And the critical variable did not seem to be time or repetition as such. What seemed to matter was the type of processing.
Their alternative was to treat memory as a by-product of perception and comprehension. When we perceive a stimulus, we analyze it through a series of stages, roughly from shallow sensory features (lines, letters, brightness) through sound to meaning and associations. Craik and Lockhart proposed that the durability of a memory trace is a function of the depth reached in that analysis. "Deeper" meant more semantic, cognitive analysis.
Two kinds of rehearsal
The framework also split rehearsal into two kinds. Maintenance rehearsal (Craik and Lockhart called it Type I processing) means recycling information at the same level, such as repeating a phone number to yourself until you dial it. Elaborative rehearsal (Type II) means processing the material more deeply, linking it to meaning and to what you already know.
The prediction was striking: maintenance rehearsal alone should do little for long-term memory. In 1973 Craik and Michael Watkins tested this. In their experiments, how long a word stayed in short-term memory and how many times it was overtly rehearsed did not predict later recall. They concluded that the maintenance and elaborative aspects of rehearsal can be separated, and that rehearsal time helps long-term learning only when it is elaborative.
Later research softened this a little. Maintenance rehearsal does slightly improve later recognition, even if it does little for free recall, so "no effect at all" overstates the case. But the main point, that sheer repetition is a poor way to learn, survived.
Earlier hints: Hyde and Jenkins
The levels idea built on experiments in which people learn words without knowing they will be tested, called incidental learning. In a well-known 1969 study, Thomas Hyde and James Jenkins gave people different "orienting tasks" for each word: some rated how pleasant each word was, others checked for particular letters. People who rated pleasantness, a task that requires thinking about meaning, later recalled many more words than those who checked letters. Remarkably, people who were told about the coming memory test did little better than people who were not, as long as they did the same orienting task. What predicted memory was the task, not the intention to remember. This was exactly the pattern Craik and Lockhart's framework explained.
The key experiments: Craik and Tulving (1975)
The most cited test of the framework is Craik and Tulving's 1975 paper in the Journal of Experimental Psychology: General, "Depth of processing and the retention of words in episodic memory." It reported ten experiments.
The basic procedure was simple. Participants saw a series of words, each preceded by a question. The questions were designed to require three different levels of analysis:
- Structural (shallow): "Is the word in capital letters?"
- Phonemic (intermediate): "Does the word rhyme with WEIGHT?"
- Semantic (deep): "Would the word fit the sentence: He met a ___ in the street?"
For each word the correct answer was "yes" for half the trials and "no" for the other half. Participants simply answered quickly. They were not told there would be a memory test. Afterward, they were given a surprise recognition or recall test.
The results lined up exactly with the framework. Memory was poorest after case questions, better after rhyme questions and best after sentence questions. In one experiment, as widely summarized in textbooks, participants correctly recognized only about 15% of words they had judged for letter case, but more than 80% of words they had judged for meaning.
Was it just more time?
An obvious objection is that semantic questions might simply take longer to answer, giving people more time with each word. Craik and Tulving addressed this. In one experiment they made the shallow task harder and slower: participants had to decide whether a word matched a particular pattern of consonants and vowels. This task took longer than the semantic judgments, but memory for those words was still worse. Time spent on a word, on its own, did not explain the effect. The kind of processing did.
The "yes" advantage
A second finding complicated the picture. Words that received a "yes" answer were generally remembered better than words that got a "no," especially in the rhyme and sentence conditions. If the word fits the sentence, it joins up with the sentence into a coherent, meaningful unit. If it does not fit, the sentence offers no structure to hold it. Craik and Tulving argued that this showed elaboration and congruity matter, not just abstract depth. A richer, better integrated encoding gives you more routes back to the memory later.
That observation pushed the framework toward a more refined idea: what matters is how elaborate and distinctive the encoding is, and how well it is integrated with what you already know.
Criticism 1: is "depth" circular?
The best-known objection, pressed by Alan Baddeley in a 1978 Psychological Review paper and by others, is that the framework had no independent measure of depth. If we judge a task as "deep" because it produces good memory, and then explain good memory by saying the task was deep, the argument goes in a circle. Processing time could not serve as an independent measure, because Craik and Tulving's own experiments showed that time and memory could come apart.
Craik and Lockhart acknowledged this weakness. Later researchers tried to anchor depth in independent measures, such as ratings of how semantic a task is or patterns of brain activity, but no universally accepted yardstick has emerged.
Criticism 2: transfer-appropriate processing
A more damaging challenge came in 1977 from C. Donald Morris, John Bransford and Jeffery Franks. They had participants encode words either semantically or by rhyme, then tested them in two different ways. On a standard recognition test ("Did you see this word?"), semantically encoded words were remembered better, as levels of processing predicts. But on a rhyming recognition test ("Did you see a word that rhymes with this one?"), words encoded by rhyme were remembered better.
This result showed that "deep" processing is not automatically superior. What matters is the match between how you encode information and what the test asks you to do. Morris and colleagues called this transfer-appropriate processing. It connects to Tulving's encoding specificity principle, which holds that retrieval works best when retrieval cues match the way the memory was originally encoded.
The two ideas now usually coexist. Semantic processing tends to win on typical memory tests because those tests ask about meaning, and because meaning-based encodings link to more cues. But on a test that requires surface detail, surface encoding can be the better choice.
Criticism 3: not all memory works this way
Levels of processing effects are strong in explicit memory tests, where people deliberately try to recall or recognize. In implicit memory tests, such as completing word fragments without trying to remember, the effect of semantic versus perceptual encoding is often much smaller or absent. Perceptual priming can depend more on surface features. Motor skills, which rely on procedural memory, also follow different rules; repeated practice is essential there.
Craik himself has stressed that the original paper was intended as a framework for research, not a complete theory. In a 2002 retrospective in the journal Memory, he reviewed thirty years of work and argued that the core claim, that memory is a by-product of processing and that deeper, more elaborate processing generally yields more durable memory, still stands, while the specifics have been revised.
Extensions: elaboration, distinctiveness and the self
The levels framework spawned research programs that refined what "deep" means.
Elaboration. Within the same level of processing, richer encoding helps. Craik and Tulving found that words judged against more complex, detailed sentences were recalled better than words judged against simple sentences, even though both tasks were semantic. Elaboration gives a memory more connections, and more connections mean more possible routes back to it.
Distinctiveness. Researchers including R. Reed Hunt argued that what makes an encoding effective is partly how distinctive it is, how much it sets an item apart from everything else in memory. Semantic processing tends to produce distinctive traces because meanings differ more than letter shapes do. But an item processed in an unusual way can be well remembered even without deep semantic analysis.
The self-reference effect. In 1977, T. B. Rogers, N. A. Kuiper and W. S. Kirker added a fourth question to the classic design: "Does this word describe you?" Words judged in relation to oneself were remembered even better than words judged for general meaning. The self is a large, well-organized body of knowledge, so relating a word to it is an especially rich form of elaboration. The self-reference effect has been replicated many times and is one of the most useful practical extensions of the framework.
Levels of processing and aging
Craik went on to apply these ideas to memory in later life, a major focus of his career at the University of Toronto and later the Rotman Research Institute at Baycrest. A recurring finding in aging research is that older adults tend to show larger memory deficits in free recall, which demands self-initiated processing, than in recognition, where the test itself provides strong cues. Craik suggested that older adults are less likely to spontaneously engage in deep, elaborative processing, and that their memory improves when tasks provide what he called environmental support, such as orienting questions that guide them toward meaning. In other words, part of the age-related decline in memory may reflect how information is processed at encoding, not only how well it is stored.
The brain basis
Neuroimaging studies have found brain signatures that track depth of processing. In a 1994 PET study, Shitij Kapur and colleagues reported greater activation of the left inferior prefrontal cortex when people made semantic judgments about words than when they made perceptual judgments, alongside better later memory for the semantically processed words.
In 1998 Anthony Wagner and colleagues used event-related fMRI to look at individual words. Words that would later be remembered had produced stronger activity during encoding in left prefrontal and medial temporal regions than words that would later be forgotten. This "subsequent memory effect" fits the idea that the quality of processing at encoding helps decide what lasts. Such findings do not prove the levels framework, but they are consistent with it.
Has the effect replicated?
Yes, robustly. The basic advantage of semantic over shallow orienting tasks has been reproduced in countless classroom demonstrations, lab studies and textbooks for five decades, and it holds across ages and many kinds of materials. It is far more reliable than many effects that became famous in social psychology. What has not held up is the strong theoretical claim that depth alone determines memory; transfer-appropriate processing, elaboration and distinctiveness all matter too.
Common misconceptions
"Craik and Lockhart's 1972 study found 80% versus 20% recall." The 1972 paper reported no new experiment, and the often-quoted round numbers do not come from it. The widely cited contrast of about 15% versus over 80% recognition is from Craik and Tulving's 1975 experiments.
"Repetition is useless." Repetition that merely recycles surface information is a weak strategy for long-term memory of facts. But repeated retrieval practice, spaced over time, is one of the most effective study methods known. The forgetting curve shows why reviews are needed at all.
"Deep always beats shallow." Only when the test depends on meaning. Transfer-appropriate processing shows the opposite can happen.
"Mnemonics work because they are deep." Some do, because they add meaning or imagery. Others, like acronyms, mainly provide a retrieval structure. They work for several reasons, not just depth.
Try it yourself
The effect is easy to demonstrate with a friend or a class. Prepare a list of about 30 common concrete nouns. For each word, write one of three questions on a card: whether the word is in capital letters, whether it rhymes with a given word, or whether it fits a given sentence. Mix the question types in random order and make half of the correct answers "yes" and half "no."
Show each question, then the word, and ask for a quick yes or no answer. Do not mention a memory test. When the list is finished, do something unrelated for a few minutes, then ask everyone to write down as many words as they can remember. Score recall separately for each question type.
With enough participants, the results almost always fall in the classic order: case questions lowest, rhyme questions in the middle and sentence questions highest, with "yes" words usually beating "no" words. Exact percentages will differ from Craik and Tulving's because of the list, the delay and the type of test, but the pattern is reliable enough that it is a standard teaching demonstration.
What it means for studying
Levels of processing helped push educators away from passive rereading and toward active engagement. Practical strategies consistent with the research include:
- Explain material in your own words. Paraphrasing forces you to work with meaning instead of surface wording.
- Ask why and how. Elaborative interrogation, generating explanations for why a fact is true, has moderate support in reviews of study techniques.
- Connect to what you know. Relating new material to personal experience or prior knowledge is a form of elaboration; the "self-reference effect," better memory for words judged in relation to oneself, is a well-replicated extension of levels of processing.
- Test yourself in the format you will need. Transfer-appropriate processing suggests practicing the kind of retrieval the exam or real task will require.
- Space your reviews. Depth slows forgetting but does not stop it.
For more on how memory goes wrong, see why we forget and the Loftus misinformation experiments. For working memory limits, see how working memory limits thinking, and for a broad overview, the science of learning.
The takeaway
Craik and Lockhart changed the question memory researchers asked. Instead of "which store is this item in, and for how long?" they asked "what did the mind do with this item?" Craik and Tulving's experiments showed how much that matters: the same one-second exposure to a word could lead to good or poor memory depending on the question asked about it. The framework's vagueness about depth and its blind spot for test conditions drew fair criticism, and transfer-appropriate processing gave it a needed correction. But the practical lesson has held for half a century: what you remember depends heavily on how you think about it.
Frequently Asked Questions
Q: What is the levels of processing theory? A: It is the idea, proposed by Fergus Craik and Robert Lockhart in 1972, that how well information is remembered depends on how deeply it is processed when it is encountered. Processing meaning produces more durable memories than processing appearance or sound.
Q: What experiment tested levels of processing? A: The best-known test is Craik and Tulving (1975). Participants answered questions about each word's letter case, rhyme or fit in a sentence, then took a surprise memory test. Memory was poorest after case questions and best after sentence questions; in one experiment recognition was about 15% versus over 80%.
Q: Did Craik and Lockhart run the famous experiment? A: No. The 1972 Craik and Lockhart paper was a theoretical framework that reviewed earlier evidence. The best-known experiments came from Craik and Endel Tulving in 1975.
Q: What is the main criticism of levels of processing? A: That "depth" was never independently defined, so the explanation can become circular. Transfer-appropriate processing (Morris, Bransford and Franks, 1977) also showed that shallow, rhyme-based encoding can beat semantic encoding when the test asks about rhymes.
Q: What is the difference between maintenance and elaborative rehearsal? A: Maintenance rehearsal repeats information at the same level, like repeating a phone number. Elaborative rehearsal connects it to meaning and prior knowledge. Craik and Watkins (1973) found that the amount of maintenance rehearsal did not predict later recall.
Q: If I process something deeply, will I remember it forever? A: No. Deep processing slows forgetting but does not stop it. Spaced review and retrieval practice are still needed for long-term retention.
Q: Does levels of processing apply to skills like playing the piano? A: Not directly. The framework was developed for explicit memory of verbal material. Motor skills depend on procedural memory, which improves with repeated, well-structured practice.
Sources
- Craik FIM, Lockhart RS (1972). Levels of processing: A framework for memory research. J Verbal Learning Verbal Behav 11:671–684
- Craik FIM, Tulving E (1975). Depth of processing and the retention of words in episodic memory. J Exp Psychol: General 104:268–294 (APA PsycNet)
- Craik FIM, Watkins MJ (1973). The role of rehearsal in short-term memory. J Verbal Learning Verbal Behav 12:599–607 (PDF)
- Morris CD, Bransford JD, Franks JJ (1977). Levels of processing versus transfer appropriate processing. J Verbal Learning Verbal Behav 16:519–533
- Atkinson RC, Shiffrin RM (1968). Human memory: A proposed system and its control processes
- Baddeley AD (1978). The trouble with levels: A reexamination of Craik and Lockhart's framework. Psychological Review 85:139–152
- Craik FIM (2002). Levels of processing: Past, present... and future? Memory 10:305–318
- Wagner AD et al. (1998). Building memories: Remembering and forgetting of verbal experiences as predicted by brain activity. Science 281:1188–1191
- Kapur S et al. (1994). Neuroanatomical correlates of encoding in episodic memory: levels of processing effect. PNAS 91:2008–2011
- Rogers TB, Kuiper NA, Kirker WS (1977). Self-reference and the encoding of personal information. JPSP 35:677–688
- American Academy of Arts & Sciences: Craik on memory processes and aging
- Kihlstrom JF, Encoding (Berkeley memory course notes)
Read next:
Follow new articles
Add SciHub101 to your feed reader to see new science and technology stories.
About the Author
SciHub101 is an independent science and technology publication founded and edited by Buğra Sözeri, a software engineer and app developer based in Türkiye. AI tools assist with drafting and research. Source documentation in the existing archive is uneven, and articles are being reviewed and corrected. The Editorial Team byline identifies editorial responsibility; it does not certify specialist or independent human review. Where we compare products, anything built by our own team is labelled as such — see the editorial guidelines and About page for how corrections and disclosures work.
Related Articles

Active Recall and Quiz-Based Learning: The Science Behind the Most Powerful Study Method
Decades of cognitive psychology research reveals that the most effective study method is not reading or highlighting. It is active recall: testing yourself on what you have learned, spacing those tests over time, and using retrieval difficulty as the signal for what to study next. The research, pioneered by Roediger and Karpicke, shows that retrieval practice produces substantially larger long-term learning gains than equivalent study time spent on relearning. This article explains the science and shows how to apply it to any subject.

Why Do We Forget? The Science of Memory Decay and Forgetting
Forgetting is not a failure of memory but an adaptive feature of the mind. Information is encoded in the brain through synaptic strengthening, stored in distributed patterns across neural networks, and retrieved by reconstructing those patterns. Forgetting occurs through two mechanisms: decay (the gradual weakening of memories over time if unused) and interference (new information competing with or overwriting old information). The forgetting curve, first measured by Hermann Ebbinghaus in 1885, shows that memories fade rapidly at first, then stabilize. Motivated forgetting — the intentional suppression of unwanted memories — involves the prefrontal cortex inhibiting memory retrieval circuits. Understanding forgetting reveals that memory loss can be adaptive: it reduces cognitive clutter and allows the mind to prioritize relevant, recent information.

Pavlov's Classical Conditioning: Learning Theory Explained
How Ivan Pavlov's dogs revolutionized understanding of learned behavior and memory formation

The Loftus Misinformation Effect: How Memory Gets Rewritten
Elizabeth Loftus's 1974 car-crash study showed a single word can alter memory. Here's how she proved it, what it means for eyewitness testimony, and why courts still struggle with the findings.