Learning

Levels of Processing Theory: Deep vs Shallow

In 1972, Fergus Craik and Robert Lockhart argued that memory depends less on how many times you go over something than on how deeply you think about what it means. Here's the evidence behind the levels of processing theory, the holes its critics found, and how to use it when you read, highlight or take notes.

Key Takeaways
    • Memory follows depth of processing: Craik and Lockhart (1972) proposed that how long a memory lasts depends on how deeply the material was analyzed, from surface features like how a word looks or sounds to what it means.
  • Meaning beats appearance by a wide margin: In the first of Craik and Tulving's 1975 experiments, people later recognized 18% of the words they had said were in capital letters but 96% of the words they had said fit a sentence, without trying to memorize any of them.
  • Repetition on its own adds little: Saying something over and over (maintenance rehearsal) didn't improve later recall in Craik and Watkins's 1973 experiments. Connecting it to what you already know does the work.
  • Depth isn't the whole story: Critics pointed out that "depth" was never measured independently of memory, and Morris, Bransford and Franks (1977) showed that the best way to encode something depends on how you'll be tested.
  • Highlighting is shallow until you add meaning: A 2013 review rated highlighting "low utility." A highlight becomes deep processing when you add a note in your own words, ask why it's true, or link it to something you know.
  • Pair deep processing with retrieval: Ask why, relate the material to yourself, explain it, then test yourself from memory in the format you'll face.

What Is the Levels of Processing Theory?

The levels of processing theory says that how well you remember something depends on how deeply you processed it when you first took it in. Shallow processing deals with surface features, such as what a word looks like or how it sounds. Deep processing deals with meaning: what the word refers to, how it connects to what you already know, and why it matters. The deeper the processing, the more durable the memory tends to be.

Fergus Craik and Robert Lockhart, two psychologists at the University of Toronto, proposed the idea in a 1972 paper in the Journal of Verbal Learning and Verbal Behavior, "Levels of processing: A framework for memory research." It went on to become one of the most cited papers in the history of cognitive psychology.

The paper took aim at the leading model of the day, the multi-store model that Richard Atkinson and Richard Shiffrin had published in 1968. That model pictured memory as a set of boxes (a sensory register, a short-term store and a long-term store), with attention and rehearsal moving information from one box to the next. You could draw a box around the early stages of analysis and call it sensory memory, and another around the intermediate stages and call it short-term memory, Craik and Lockhart wrote, "but that procedure both oversimplifies matters and evades the more significant issues."

Instead of asking where a memory is kept, they asked what the person did with the information when it arrived. Their central claim fits in one sentence: "trace persistence is a function of depth of analysis, with deeper levels of analysis associated with more elaborate, longer lasting, and stronger traces."

On this view, memory is a by-product of perceiving and understanding. You remember the gist of a conversation you found interesting without trying to, and you forget a phone number you repeated twenty times, because repeating it never required you to understand anything.

They called it a "framework" and meant it: "Our approach does not constitute a theory of memory," they wrote. That looseness, above all the lack of any independent measure of depth, became the critics' main target.


Structural, Phonemic and Semantic: The Three Levels of Processing

Textbooks usually describe three levels, from shallowest to deepest. The labels come mostly from the experiments that followed the 1972 paper. Craik and Lockhart themselves wrote that processing levels "may be more usefully envisaged as a continuum of analysis," so treat these as points on a scale rather than three fixed steps.

LevelWhat you attend toExample question from the experimentsWhat it looks like when you study
Structural (visual)How it looks: letters, typeface, capitals"Is the word in capital letters?"Noticing that a term is in bold without reading its definition
Phonemic (acoustic)How it sounds: rhyme, pronunciation"Does the word rhyme with WEIGHT?"Repeating a definition aloud until it sounds right
SemanticWhat it means and how it connects to other ideas"Would the word fit the sentence 'He met a ___ in the street'?"Explaining the term in your own words, with your own example

Structural processing is the shallowest. You can tell whether a word is in capitals without knowing what it means, so little of it sticks. Phonemic processing goes a step further: to judge a rhyme you have to hear the word in your head, but you still don't have to understand it. Semantic processing is the deepest, because you can't decide whether a word fits a sentence without thinking about what it means.

Even semantic processing comes in degrees. By 1975, Craik and Tulving were already talking about "degrees of stimulus elaboration," meaning how much a word picks up while you process it. In one of their experiments, words placed in complex, detailed sentences were recalled better than words placed in simple ones, as long as the word fit. A word linked to one fact is processed semantically. A word linked to a vivid scene, a personal memory and a related idea is processed semantically and richly, and it has more routes back into memory.


The Key Experiments Behind Levels of Processing

The 1972 paper was mostly argument, built partly on a few earlier studies. Most of the evidence came in the years that followed, and the most convincing of it came from experiments in which people weren't trying to memorize anything.

Craik and Tulving (1975): meaning beats appearance. Fergus Craik and Endel Tulving ran ten experiments that became the standard demonstration. In the first, participants heard a question about a word, saw the word flashed for 200 milliseconds, and answered: was it in capitals, did it rhyme with another word, did it belong to a category, did it fit a sentence? They were told the study was about perception and reaction time. Then came a recognition test nobody had mentioned.

For questions answered "yes," participants recognized 18% of the words they had judged for capitals, 78% of those judged for rhyme and 96% of those fitted into a sentence. The deeper questions took only a fraction of a second longer to answer. Counting that extra decision time as study time, the authors noted, "200 msec of extra study time led to a 400% improvement in retention."

Hyde and Jenkins (1969, 1973): the task matters more than the intention. Thomas Hyde and James Jenkins gave people word lists and a job to do with each word. Some rated how pleasant each word was, which requires thinking about its meaning. Others checked the words for particular letters or estimated how many letters they had. Some groups were warned of a memory test and others weren't. Rating pleasantness, as Craik and Lockhart summarized it, "produces performance equal to that of intentional learning," and both beat the letter tasks substantially. (A 2023 study by Vencislav Popov and Hannah Dames argues that the intent to learn does help, an effect the classic designs tended to hide. Their participants all processed words for meaning, so intent adds to deep processing rather than replacing it.)

Rogers, Kuiper and Kirker (1977): the self-reference effect. Timothy Rogers and his two co-authors added a fourth question to the usual three: "Describes you?" Adjectives judged that way were recalled better than adjectives judged for appearance, sound or even meaning. In their first experiment, recall was about twice as high as in the meaning condition. Twenty years later, a meta-analysis of 129 studies by Cynthia Symons and Blair Johnson found that the effect held up: relating material to yourself beat both ordinary semantic processing and relating it to someone else.

Kapur and colleagues (1994): depth shows up in the brain. Shitij Kapur, Craik, Tulving and three colleagues used PET scans while volunteers judged words either as living or non-living, or for whether they contained the letter "a." The meaning task led to substantially better recognition afterward and to more activity in the left inferior prefrontal cortex.


Maintenance vs Elaborative Rehearsal: Why Repetition Isn't Enough

Craik and Lockhart separated two kinds of rehearsal, which they called Type I and Type II processing. Maintenance rehearsal (Type I) is "repetition of analyses which have already been carried out," like saying a phone number over and over until you dial it. Elaborative rehearsal (Type II) "involves deeper analysis of the stimulus," linking it to meaning, to other facts or to yourself. Their prediction was blunt: "Only this second type of rehearsal should lead to improved memory performance."

Fergus Craik and Michael Watkins tested it in 1973. Neither how long a word was held in mind nor how often it was rehearsed aloud predicted later recall. In one experiment, students rehearsed the last four words of each list about six extra times per word during a 20-second delay. On a surprise final test, they recalled 9.1% of those words, slightly less than the 9.8% they recalled when there was no delay and no extra rehearsal. The authors' conclusion: "Maintenance rehearsal does not lead to an improvement in memory performance." A 1977 study by Arthur Glenberg, Steven Smith and Carol Green added a footnote: nine times as much rote rehearsal barely moved recall, but it did lift recognition a little. Repetition can make things feel familiar without making them easier to recall.

For students, maintenance rehearsal hides in habits that feel productive:

  • Rereading a chapter until it looks familiar.
  • Reciting a definition word for word.
  • Copying slides or notes neatly without changing a word.
  • Rewatching a lecture start to finish at the same pace.

Each keeps the material in front of you, and each can leave you feeling you know it. That feeling is familiarity, and it often collapses when an exam asks you to use the idea. Elaborative rehearsal looks different: asking how a new term relates to one you know, building a sentence of your own around it, or explaining why a result came out the way it did. It's slower and harder, so it rarely feels as productive as it is.


Criticisms of the Levels of Processing Theory

The framework changed memory research, but it took real hits. Four criticisms come up most often.

1. "Depth" is circular. There's no independent ruler for depth. If deep processing means whatever leads to good memory, then "deep processing leads to good memory" can't be wrong, and a claim that can't be wrong doesn't explain much. Craik and Tulving saw the danger themselves, warning that "any well-remembered event can too easily be labeled deeply processed." Alan Baddeley pressed the point in a 1978 critique titled "The trouble with levels," and Michael Eysenck made the same complaint that year: "there are no suitable criteria available for indexing either the depth or the spread of encoding."

2. The best encoding depends on the test. In 1977, Donald Morris, John Bransford and Jeffery Franks had people process words either for meaning or for rhyme. On a standard recognition test, meaning won as usual, about 84% against 63%. On a test asking which new words rhymed with the studied ones, the order flipped: about 49% for rhyme processing against 33% for meaning. They called it transfer-appropriate processing: memory depends on the match between how you processed something and what the test asks you to do with it. (The 1972 paper had hinted as much.)

3. Deeper tasks usually take longer. Maybe time, not depth, explains the advantage. Craik and Tulving tested this with a shallow task that was slow and hard: deciding whether a word matched a pattern of consonants and vowels, like CCVVC for "brain." It took about twice as long as a quick sentence judgment (1.7 seconds against 0.85), yet the sentence words were recognized more often (82% against 57% for "yes" answers). Effort and time on task aren't the same as depth.

4. It describes more than it explains. Why should meaning last longer? Later work explained the results through elaboration (how richly an item is encoded) and distinctiveness (how much it stands out from other memories), and Craik's 2002 retrospective took up the common criticisms directly, from measuring depth to the interplay of encoding and retrieval.

The core finding survives: for the kind of memory most courses test, meaning beats surface features. But your studying should also resemble the way you'll use the knowledge later.


Shallow vs Deep Processing: Examples From Everyday Studying

Most study habits come in a shallow version and a deep version.

ActivityShallow versionDeep version
Reading a textbook chapterReading straight through, then rereading the bold termsPausing at each claim to ask why it's true and how it connects to the last one
HighlightingMarking whatever looks important as you goReading the paragraph first, marking one key sentence and noting why it matters
Taking lecture notesCopying the slides word for wordWriting each point in your own words, with an example
Learning vocabularyRepeating a word and its definitionWriting your own sentence that uses the word
Using flashcardsFlipping the card as soon as the answer feels familiarAnswering out loud before you flip, then explaining why the answer is right
Watching a recorded lecturePlaying it at 2x with no pausesPausing every few minutes to summarize from memory what you just heard

Note-taking research makes a similar argument. In 2014, Pam Mueller and Daniel Oppenheimer reported that students who typed notes on laptops did worse on conceptual questions than students who wrote by hand, and they blamed "shallower processing": typists tended to transcribe lectures verbatim "rather than processing information and reframing it in their own words." The laptop part hasn't held up well. A 2019 direct replication by Kayla Morehead, John Dunlosky and Katherine Rawson found that performance didn't consistently differ between groups. The processing part is the useful bit: copying is shallow whatever you copy with. Our guide to handwriting vs typing notes covers those studies in detail.


Is Highlighting Shallow Processing?

Usually, yes. Highlighting the way most people do it can happen almost entirely at the structural level. A sentence looks important because it's a definition, it's in bold or it opens a paragraph, so you mark it without working out what it means.

The research reflects that. A 2013 review of ten study techniques by John Dunlosky and four colleagues rated highlighting and underlining "low utility," alongside rereading and summarization; only practice testing and distributed practice earned "high utility." A 2022 meta-analysis by Héctor Ponce, Richard Mayer and Ester Méndez, pooling 85 effect sizes from 36 articles, found a split: highlighting that readers did themselves improved memory for the text (an effect size of about 0.36) but not comprehension (about 0.20, not a statistically reliable gain).

The levels of processing view fits both results. Marking a line singles it out, which can help you remember it, but understanding comes from working out what it means. A highlight records a decision about what matters, and you can make that decision shallowly or deeply:

  • Read first, mark second. Finish the paragraph before you highlight, so you choose based on what it says rather than how it looks.
  • Mark less. One key sentence per section forces you to judge importance, which is semantic work.
  • Add a note in your own words. Write why the passage matters or what it reminds you of. With Glasp's web highlighter, you can highlight web pages and PDFs and attach a note to any highlight, so the reason stays next to the line.
  • Ask a question in the margin. "Why would this be true?" or "Does this contradict chapter 2?" turns a mark into a prompt.
  • Review by recalling, not rereading. Cover the passage and try to explain it from the highlight alone.

For the full research on when highlighting helps and when it hurts, see our deep dive on the science of highlighting.


How to Study With Deep Processing: 7 Techniques

Each technique below forces you to work with meaning, and several have strong evidence of their own.

1. Ask "why is this true?" Turning facts into why-questions is called elaborative interrogation. In a 1987 study by Michael Pressley and colleagues, undergraduates read sentences such as "The hungry man got into the car." One group explained why that man did that, a second was handed an explanation ("to go to the restaurant") and a third just read. Asked later which man had done what, the why-question group was right about 72% of the time across the experiments, against about 37% for each of the other two groups. Reading a good explanation didn't help. Producing one did.

2. Put it in your own words. Information you produce yourself is remembered better than information you only read, a finding called the generation effect. A 2007 meta-analysis by Sharon Bertsch and colleagues pooled 445 effect sizes from 86 studies and found an average benefit of 0.40, "almost half a standard deviation" in favor of generating over reading. Paraphrase definitions, summarize sections from memory, and answer your own questions before checking the text.

3. Relate it to yourself. The self-reference effect is one of the sturdiest findings in this literature, so use it on purpose. When have you seen confirmation bias in your own decisions? Which of your habits fits the definition of negative reinforcement? You're tying new material to the most detailed body of knowledge you have.

4. Make up your own example. A definition you can recite is shallow. An example you invent has to fit the meaning, so producing one forces semantic processing and shows you quickly whether you understood. If you can't come up with one, that's useful information too.

5. Connect it to what you already know. Meaning needs something to attach to. In 1972, John Bransford and Marcia Johnson read students a paragraph that began, "The procedure is actually quite simple. First you arrange things into different groups depending on their makeup." Without a topic, it's close to gibberish. Students told beforehand that it was about washing clothes recalled 5.83 of its 18 ideas on average, about twice as many as students given no topic (2.82) or given it afterward (2.65). Before you read a chapter, skim the headings and ask what you already know about the subject.

6. Explain it to someone else. Explaining forces you to organize ideas, fill gaps and use your own words. Knowing in advance that you'll have to teach the material changes how you study it, an idea known as the protégé effect.

7. Then test yourself in the format you'll face. Deep encoding gets material in; retrieval keeps it there. In a 2006 experiment, Henry Roediger and Jeffrey Karpicke had 120 college students read two short passages, restudy one and take a recall test on the other. Students given the final test five minutes later recalled more of the restudied passage (81% of its ideas against 75%). Students tested a week later recalled more of the tested one (56% against 42%). Transfer-appropriate processing adds a rule of thumb: practice retrieving the material the way the exam will ask for it. Our guide to active recall shows how.

A few tools make the deep version easier to keep up. For lectures on YouTube, YouTube Summary shows the full transcript next to the video, so you can pause, write a summary from memory and check it against what was said. Your Kindle highlights can sit in the same place as your web highlights and notes. And Glasp's AI chat answers from your own saved highlights, so you can ask it to quiz you on what you read rather than summarize it for you.


Frequently Asked Questions

What is the levels of processing theory in simple terms?

It says that the more deeply you think about information when you first meet it, the better you'll remember it. Thinking about meaning (deep processing) leads to better memory than paying attention to how something looks or sounds (shallow processing). Fergus Craik and Robert Lockhart proposed it in 1972.

What are the three levels of processing?

Structural (how a word looks), phonemic (how it sounds) and semantic (what it means). Structural is the shallowest and semantic the deepest. Craik and Lockhart described depth as a continuum, so the three levels are convenient labels rather than fixed stages.

What is an example of deep processing?

Explaining a concept in your own words, connecting it to something you already know, or asking yourself why it's true. So is deciding whether a word describes you, which produced especially strong memory in experiments on the self-reference effect.

What is the difference between shallow and deep processing?

Shallow processing focuses on surface features, such as the letters, typeface or sound of a word, and tends to produce weak memories. Deep processing focuses on meaning and connections and produces stronger, longer-lasting ones. Rereading and repeating are usually shallow; explaining and questioning are usually deep.

How is the levels of processing model different from the multi-store model?

The multi-store model (Atkinson and Shiffrin, 1968) describes memory as separate stores, with rehearsal moving information from short-term to long-term memory. The levels of processing framework says memory depends on the kind of processing information receives, not on which store it sits in, and that repetition without deeper processing does little for long-term memory.

What are the main criticisms of the levels of processing theory?

The biggest is circularity: there's no independent measure of depth, so deep processing risks being defined as whatever is remembered well. Critics also showed that the best encoding depends on how memory is tested (transfer-appropriate processing), and that the framework describes the effect better than it explains it.

Is highlighting deep or shallow processing?

Usually shallow, because you can mark a sentence based on how it looks without thinking about what it means. It gets deeper when you read before you mark, highlight selectively and add a note about why the passage matters.


Conclusion: Think About Meaning, Then Test Yourself

More than fifty years after Craik and Lockhart's paper, the core idea has held up better than the framework around it. What you do with information when you meet it shapes what you can remember later. Looking at words, repeating them and marking them feel like studying, but they mostly work at the surface. Asking why, connecting, explaining and relating ideas to yourself work on meaning, and meaning is what lasts.

The debates that followed supplied the other half of the lesson: depth isn't the same as effort, and the best preparation looks like the test. So process deeply when you learn, then practice retrieving what you learned.

Glasp is built for that loop: highlight on the web and in PDFs, import your Kindle highlights, jot down why each passage matters, and come back to quiz yourself instead of rereading.


References

Atkinson, R. C., and Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. Psychology of Learning and Motivation, 2, 89-195. doi: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. doi:10.1037/0033-295X.85.3.139

Bertsch, S., Pesta, B. J., Wiscott, R., and McDaniel, M. A. (2007). The generation effect: A meta-analytic review. Memory and Cognition, 35(2), 201-210. doi:10.3758/BF03193441

Bransford, J. D., and Johnson, M. K. (1972). Contextual prerequisites for understanding: Some investigations of comprehension and recall. Journal of Verbal Learning and Verbal Behavior, 11(6), 717-726. doi:10.1016/S0022-5371(72)80006-9

Craik, F. I. M., and Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11(6), 671-684. doi:10.1016/S0022-5371(72)80001-X

Craik, F. I. M., and Tulving, E. (1975). Depth of processing and the retention of words in episodic memory. Journal of Experimental Psychology: General, 104(3), 268-294. doi:10.1037/0096-3445.104.3.268

Craik, F. I. M., and Watkins, M. J. (1973). The role of rehearsal in short-term memory. Journal of Verbal Learning and Verbal Behavior, 12(6), 599-607. doi:10.1016/S0022-5371(73)80039-8

Craik, F. I. M. (2002). Levels of processing: Past, present... and future? Memory, 10(5-6), 305-318. doi:10.1080/09658210244000135

Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., and Willingham, D. T. (2013). Improving students' learning with effective learning techniques: Promising directions from cognitive and educational psychology. Psychological Science in the Public Interest, 14(1), 4-58. doi:10.1177/1529100612453266

Eysenck, M. W. (1978). Levels of processing: A critique. British Journal of Psychology, 69(2), 157-169. doi:10.1111/j.2044-8295.1978.tb01643.x

Glenberg, A., Smith, S. M., and Green, C. (1977). Type I rehearsal: Maintenance and more. Journal of Verbal Learning and Verbal Behavior, 16(3), 339-352. doi:10.1016/S0022-5371(77)80055-8

Hyde, T. S., and Jenkins, J. J. (1969). Differential effects of incidental tasks on the organization of recall of a list of highly associated words. Journal of Experimental Psychology, 82(3), 472-481. doi:10.1037/h0028372

Hyde, T. S., and 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. doi:10.1016/S0022-5371(73)80027-1

Kapur, S., Craik, F. I. M., Tulving, E., Wilson, A. A., Houle, S., and Brown, G. M. (1994). Neuroanatomical correlates of encoding in episodic memory: Levels of processing effect. Proceedings of the National Academy of Sciences, 91(6), 2008-2011. doi:10.1073/pnas.91.6.2008

Morehead, K., Dunlosky, J., and Rawson, K. A. (2019). How much mightier is the pen than the keyboard for note-taking? A replication and extension of Mueller and Oppenheimer (2014). Educational Psychology Review, 31(3), 753-780. doi:10.1007/s10648-019-09468-2

Morris, C. D., Bransford, J. D., and Franks, J. J. (1977). Levels of processing versus transfer appropriate processing. Journal of Verbal Learning and Verbal Behavior, 16(5), 519-533. doi:10.1016/S0022-5371(77)80016-9

Mueller, P. A., and Oppenheimer, D. M. (2014). The pen is mightier than the keyboard: Advantages of longhand over laptop note taking. Psychological Science, 25(6), 1159-1168. doi:10.1177/0956797614524581

Ponce, H. R., Mayer, R. E., and Méndez, E. E. (2022). Effects of learner-generated highlighting and instructor-provided highlighting on learning from text: A meta-analysis. Educational Psychology Review, 34(2), 989-1024. doi:10.1007/s10648-021-09654-1

Popov, V., and Dames, H. (2023). Intent matters: Resolving the intentional versus incidental learning paradox in episodic long-term memory. Journal of Experimental Psychology: General, 152(1), 268-300. doi:10.1037/xge0001272

Pressley, M., McDaniel, M. A., Turnure, J. E., Wood, E., and Ahmad, M. (1987). Generation and precision of elaboration: Effects on intentional and incidental learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13(2), 291-300. doi:10.1037/0278-7393.13.2.291

Roediger, H. L., and Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249-255. doi:10.1111/j.1467-9280.2006.01693.x

Rogers, T. B., Kuiper, N. A., and Kirker, W. S. (1977). Self-reference and the encoding of personal information. Journal of Personality and Social Psychology, 35(9), 677-688. doi:10.1037/0022-3514.35.9.677

Symons, C. S., and Johnson, B. T. (1997). The self-reference effect in memory: A meta-analysis. Psychological Bulletin, 121(3), 371-394. doi:10.1037/0033-2909.121.3.371

Start building your knowledge library

Highlight what matters as you read across the web. Save insights from articles, books, and YouTube videos in one place.

Get Started Free

Or highlight this page as you read it