Metacognition

Metacognition is the mind keeping watch on itself. This page explains how people sense what they know, predict what they will remember, judge how well a task is going, and step in to change course, and why those inner readings are often useful but rarely exact. Three interactive tasks let you measure your own monitoring as you read.

Metacognition is cognition about cognition: the knowledge people hold about their own mental processes together with the monitoring and control that let them regulate those processes in flight (Flavell, 1979). It is what tells you that a passage has stopped making sense, that a name is on the verge of arriving, that you have studied enough, or that your current approach to a problem is going nowhere. A central and slightly unsettling finding organizes the whole field: these self-assessments are not direct read-outs of an internal record but inferences drawn from cues such as how fluently material is processed or how familiar a question feels, which makes them serviceable much of the time and systematically wrong in predictable ways (Koriat, 1997; Fleming, 2024). The sections below build the architecture from the ground up: first what metacognition is and how monitoring and control fit together, then the specific judgments people make about memory, then the calibration of confidence, and finally the neural and developmental basis of the capacity.

What Metacognition Is

The term was coined in the 1970s to name a layer of cognition that takes other cognition as its object (Flavell, 1979). A first-order process perceives a scene, retrieves a word, or solves an equation; a metacognitive process represents that first-order process and evaluates or steers it. The distinction matters because the two can dissociate: a person can remember well yet judge their memory poorly, or reason correctly yet feel uncertain, which means the monitoring system is a separate mechanism with its own accuracy to account for.

A durable taxonomy splits the field in two (Schraw & Moshman, 1995). Metacognitive knowledge is relatively stable, statable belief about cognition, and Flavell divided it into three kinds: person knowledge (beliefs about oneself and people in general as thinkers, such as believing one learns better by listening than by reading), task knowledge (how task demands shape difficulty, such as knowing that recall is harder than recognition), and strategy knowledge (which procedures serve which goals) (Flavell, 1979). Metacognitive regulation is the activity that runs during a task: planning an approach, monitoring progress, and evaluating the result. Knowledge feeds regulation and regulation updates knowledge, so the two are interdependent rather than separate stores.

The Monitoring and Control Architecture

The most influential framework casts metacognition as traffic between two levels (Nelson & Narens, 1990). The object level carries out cognition itself. The meta level holds a dynamic model of the object level and stands in two relations to it. Monitoring is the upward flow: the meta level is informed by the state of the object level, as when a feeling of difficulty signals that encoding is going badly. Control is the downward flow: the meta level modifies the object level, as when that feeling prompts you to slow down, reread, or switch strategies. Crucially the two are asymmetric and complementary, because control without monitoring is blind and monitoring without control is inert; effective self-regulation needs the loop to close. Figure 1 shows the arrangement.

This two-level picture has proven durable partly because it maps onto the brain: the same hierarchy, with a first-order process and a second-order commentary about it, can be placed on a hierarchical neural structure in which higher regions evaluate the operations of lower ones (Fleming & Dolan, 2012).

Metacognitive Knowledge

Because metacognitive knowledge guides which strategies you deploy, errors in it propagate. A student who believes that rereading is an effective way to learn will reread, and the belief is widespread even though the evidence favors retrieval practice (Dunlosky, Rawson, Marsh, Nathan, & Willingham, 2013). Person knowledge can be wrong in both directions, as when someone underestimates their capacity in an unfamiliar domain or overestimates it in a familiar one. Task knowledge improves with experience: recognizing that interleaved practice feels harder yet pays off later, or that a passage dense with new terms will need slower reading, lets a learner budget effort in advance. The practical upshot is that teaching accurate metacognitive knowledge, not just content, changes behavior, because the learner who knows which strategies work and which task features predict difficulty plans differently from one who does not (Schraw & Moshman, 1995).

Metamemory: Judgments of Learning and the Feeling of Knowing

Metamemory, metacognition about memory, is the most thoroughly mapped corner of the field because its judgments can be elicited cleanly and scored against later performance. Three judgments recur. A judgment of learning (JOL) is a prediction, made during or after study, of how likely an item is to be recalled on a later test. A feeling-of-knowing (FOK) judgment is a prediction, made after a retrieval failure, of how likely the unrecalled item is to be recognized if seen (Hart, 1965). A confidence judgment is a retrospective rating of how likely a produced answer is to be correct. Each is moderately accurate, and each is built from inference rather than direct inspection.

The accessibility account explains how the feeling of knowing is computed without privileged access to memory (Koriat, 1993). When you fail to retrieve a target, you nonetheless retrieve fragments around it, partial spellings, related facts, contextual details, and the sheer quantity and ease of that partial information drives the feeling that the answer is there. This is why the feeling is usually right, since accessible fragments tend to accompany retrievable targets, and also why it can mislead, since plausible but incorrect fragments inflate it just as well. Judgments of learning work on the same principle of cue utilization: rather than reading a memory strength meter, the learner infers future recall from cues available at the moment of judging, chiefly how fluently the item is processed (Koriat, 1997). The first demonstration lets you generate your own calibration curve for JOLs.

The Tip-of-the-Tongue State

The most vivid evidence that monitoring can access a memory it cannot retrieve is the tip-of-the-tongue state: the strong, often correct conviction that you know a word you cannot currently produce, frequently accompanied by partial information such as the first letter, the number of syllables, or words of similar sound or meaning (Brown & McNeill, 1966). The state shows that retrieval is not all-or-none and that the metacognitive feeling operates somewhat independently of the retrieval machinery, reporting on the contents of memory even when the gate to those contents is stuck. It is more frequent for proper names and low-frequency words and becomes more common with age, and its eventual resolution, often when attention has moved elsewhere, fits the picture of a target whose activation was high but momentarily insufficient.

Cue Utilization and the Illusion of Knowing

If judgments rest on cues rather than on the thing itself, then any manipulation that changes a cue without changing learning will drive the judgment astray, and several do. Retrieval fluency is the clearest case: material that comes to mind easily at the moment of judgment feels well learned, so studying a target right before rating it, or rating it while it is still in mind, inflates the judgment relative to actual durable learning (Benjamin, Bjork, & Schwartz, 1998). The same logic produces illusions of competence during study, where conditions that make performance feel easy, such as rereading a passage or studying with the answer in view, raise confidence while doing little for later recall, and conditions that make study feel effortful, such as testing yourself, do the opposite (Koriat & Bjork, 2005). The unsettling implication is that the subjective sense of knowing is a poor guide precisely where it matters most, because the cues that drive it are decoupled from the cues that predict the test. The next demonstration isolates the feeling of knowing on deliberately content-free material: after studying a handful of arbitrary pairs, then for any you cannot recall, rate whether you would recognize the answer, and find out.

Calibration and Confidence

Calibration is the correspondence between confidence and accuracy: a well-calibrated person is right about 80 percent of the time on the answers they hold with 80 percent confidence. Miscalibration takes two forms, overconfidence and underconfidence, and a robust regularity governs which appears. The hard–easy effect is the tendency to be overconfident on difficult items and underconfident on easy ones, so the gap between belief and reality widens exactly as a domain gets harder and the cues grow less diagnostic (Fleming, 2024). Calibration matters because confidence drives action: it determines when you stop studying, when you check an answer, when you seek a second opinion, and when you act unhesitatingly, so a confidence signal that runs ahead of accuracy produces premature stopping and unchecked error. The following demonstration plots your own confidence against your accuracy across a short, content-free judgment task in which difficulty is deliberately varied.

The Dunning–Kruger Effect

The best-known claim about calibration is that the least skilled are the most overconfident. In the original studies, participants in the bottom quartile on tests of humor, grammar, and logic grossly overestimated their performance, and the proposed explanation was a double burden: the skills needed to perform a task well are the same skills needed to judge that performance, so people who lack them are deprived of the very means of recognizing that they lack them (Kruger & Dunning, 1999). On this metacognitive reading, improving people's skill improves their self-assessment, which the original work reported.

The effect is genuinely contested, and an even-handed account has to say so. A statistical critique argues that much of the canonical pattern follows from regression to the mean combined with the arithmetic of comparing self-estimates to performance, so that better-than-average and worse-than-average effects would appear even without any true metacognitive deficit, and that proper individual-differences methods shrink the effect considerably (Gignac & Zajenkowski, 2020). Defenders respond that the misjudgments are real whatever their statistical scaffolding and that the metacognitive interpretation still accounts for findings a pure artifact cannot (Dunning, 2011). The defensible summary is that poor performers do tend to overestimate themselves, that this is consequential, and that how much of it reflects a specific metacognitive failure as opposed to a statistical inevitability remains under debate.

Control: Allocating Study Time

Monitoring earns its keep by feeding control, and the clearest laboratory case is the allocation of study time. Given the choice, learners do not spread effort evenly; they direct it according to their judgments of what they have and have not learned. The region of proximal learning framework holds that effective self-regulated study targets material that is not yet learned but is close to being learnable, the zone where effort yields the most return, and that learners shift away from items already mastered and away from items currently beyond reach (Metcalfe, 2009). This control policy is only as good as the monitoring behind it, which is why the timing of judgments matters so much. The delayed-JOL effect is the finding that judgments of learning made after a delay, rather than immediately after study, predict later recall far more accurately, because a delay forces the judgment to draw on durable long-term memory rather than on the fading contents of working memory (Nelson & Dunlosky, 1991). A learner who judges immediately is fooled by transient fluency; one who waits is reading the cue that actually predicts the test. Expert problem solvers show the same monitoring-and-control discipline outside of memory, devoting substantial time to planning, tracking progress, and abandoning unproductive approaches, where novices often pursue a single line without pause (Schoenfeld, 1985).

The Neural Basis

Metacognitive accuracy is dissociable from first-order ability, and the dissociation has a neural signature. Convergent evidence implicates the anterior and rostrolateral prefrontal cortex, regions disproportionately expanded in humans, in the accuracy of judgments about one's own performance, working alongside cingulate and insular cortex that track uncertainty and conflict (Fleming & Dolan, 2012). Individual differences are visible in structure as well as function: people with better metacognitive accuracy, holding task performance constant, tend to have greater gray matter volume in anterior prefrontal cortex and stronger associated white-matter integrity (Fleming, Weil, Nagy, Dolan, & Rees, 2010). Damage to prefrontal regions can impair the accuracy of self-monitoring while leaving the monitored abilities, memory and perception, comparatively intact, which is the lesion-level expression of the same separation. A current synthesis frames these judgments as propositional and model-based, the brain reasoning about its own performance using a model that can itself be more or less accurate, which is exactly why metacognition is inferential and fallible (Fleming, 2024).

Key Structures

Metacognitive monitoring and control draw on a network of higher-association regions that sit above the sensory and memory systems they evaluate. The structures below are those most consistently implicated in judging and regulating one's own cognition.

• Anterior prefrontal cortex (rostrolateral or frontopolar cortex, Brodmann area 10) — The region most reliably tied to metacognitive accuracy; greater gray-matter volume here predicts sharper monitoring independent of first-order performance (Fleming et al., 2010).
• Dorsolateral prefrontal cortex — Supports the control side of the loop, holding goals and task models in mind and adjusting strategy and study allocation in response to monitoring signals.
• Medial prefrontal cortex — Engaged in self-referential evaluation and in computing confidence for memory and value-based decisions.
• Anterior cingulate cortex — Tracks conflict, error, and uncertainty, supplying the signals that tell the meta level when something has gone wrong.
• Anterior insula — Represents interoceptive and uncertainty signals that contribute to subjective feelings of confidence and doubt.
• Precuneus — A medial parietal region recruited during self-referential metacognitive judgments and the retrieval of the episodic detail that feeds them.

The Development of Metacognition

Metacognition is not present in mature form from the start; it develops across childhood and adolescence (Flavell, 1979). Preschoolers show fragments of it, distinguishing easy from hard tasks and grasping that more items are harder to remember, yet their monitoring is poorly calibrated and they routinely overestimate what they have learned. Across the elementary years calibration improves and children begin to let monitoring guide study, and the largest gains in deliberate metacognitive regulation arrive in adolescence, tracking the protracted maturation of prefrontal cortex. The capacity never becomes uniform: even in adults, metacognition is more accurate in domains of expertise, where cues are diagnostic and feedback has been plentiful, than in unfamiliar ones, so the developmental story continues, domain by domain, throughout life.

Comparing the Three Metamemory Judgments

Table 1 sets the three judgments side by side, separating what each estimates from the cue it leans on and the error it characteristically makes.

Note. All three judgments are inferential rather than direct read-outs of memory strength, which is why each can be driven off course by manipulating its cue without changing what was actually learned (Koriat, 1997).

Worked Example

Renah is studying foreign-language vocabulary for an exam. As she reads each pair, fluent ones, words that resemble English, slide by easily and feel learned, so her immediate judgments of learning run high; the feeling is a cue-driven inference from fluency, not a reading of durable memory, and it is exactly the cue that a later test will not honor (Koriat, 1997; Benjamin, Bjork, & Schwartz, 1998). Sensing that even spacing is wasteful, she concentrates effort on pairs that are not yet learned but feel close, steering toward her region of proximal learning and away from pairs already mastered or hopelessly opaque (Metcalfe, 2009). Late in the session she quizzes herself on an earlier batch rather than judging it on the spot, and the delay exposes which items have actually held, sharpening both her judgment and her memory, a small instance of the delayed-JOL effect doing its work (Nelson & Dunlosky, 1991). On a topic she finds easy she is mildly overconfident and nearly stops too soon, the hard–easy effect in miniature, until a missed retrieval recalibrates her (Fleming, 2024). Throughout, the episode is the monitoring-and-control loop turning: fluency, accessibility, and test outcomes flow up as monitoring, and decisions about what to restudy flow down as control (Nelson & Narens, 1990).

Why It Matters

Metacognition is load-bearing wherever learning and judgment have consequences. In education it is among the strongest levers available: students who monitor accurately and direct study accordingly outperform equally able peers who do not, and the techniques that most improve learning, chiefly retrieval practice and spaced study, work in part by correcting the monitoring that governs study decisions (Dunlosky, Rawson, Marsh, Nathan, & Willingham, 2013). The same machinery explains characteristic failures: learners who overestimate their grasp stop studying too soon, and the consequences of miscalibration extend well beyond the classroom, as when an eyewitness who expresses high confidence sways a jury regardless of accuracy, or a professional acts on a judgment whose confidence outruns its evidence (Kruger & Dunning, 1999). The most reliable remedy is not exhortation to be more accurate but redesign of the procedure, since the inferential cues that drive monitoring arrive unbidden, so inserting delayed judgments, retrieval practice, and explicit checks at known failure points recalibrates self-assessment where willpower cannot (Nelson & Dunlosky, 1991). For the individual reader, the practical skill is to distrust the feeling of knowing exactly where its cues are least diagnostic, on unfamiliar material, on fluently-read text, and on answers that arrive too easily, and to test rather than reread. The demonstrations above are small drills in that habit.

Key Researchers

John H. Flavell (1928–2025). Coined the term metacognition and proposed its first formal model, dividing the field into metacognitive knowledge, experiences, goals, and strategies; Anne T. and Robert M. Bass Professor of Psychology, Emeritus, at Stanford University.

Thomas O. Nelson (1944–2005). With Louis Narens of the University of California, Irvine, formalized the monitoring-and-control framework that organizes much of the field, distinguishing the object level from the meta level.

Stephen M. Fleming. Professor of Cognitive Neuroscience at University College London; maps the computational and neural basis of metacognition and confidence. MetaLab · ORCID.

Asher Koriat. Professor at the University of Haifa; developed the accessibility model of the feeling of knowing and the cue-utilization account of judgments of learning. University of Haifa (IIPDM) · Google Scholar.

Janet Metcalfe. Professor of Psychology at Columbia University; developed the region-of-proximal-learning framework for the control of study. Columbia faculty page · Google Scholar.

David Dunning. Mary Ann and Charles R. Walgreen, Jr., Professor at the University of Michigan; with Justin Kruger, described the relationship between skill and self-insight that bears their names. Michigan faculty page.

Key Terms

Frequently Asked Questions

What is metacognition?
Metacognition is cognition about cognition: the knowledge people hold about their own mental processes together with the monitoring and control that let them regulate those processes while a task is underway. It is what lets you notice you have stopped understanding, judge whether you have studied enough, or change strategy when one is failing (Flavell, 1979).

What is the difference between metacognitive knowledge and metacognitive regulation?
Metacognitive knowledge is relatively stable belief about how cognition works, divided into beliefs about yourself as a thinker, about task demands, and about which strategies serve which goals. Metacognitive regulation is the live activity of planning, monitoring, and evaluating a process as it runs. Knowledge informs regulation and regulation updates knowledge (Schraw & Moshman, 1995).

What is a judgment of learning?
A judgment of learning is a prediction of how likely you are to recall something later, made during or just after studying it. Such judgments are inferred from cues such as how fluently the material is processed rather than read directly from memory, which is why easy-to-read material can feel learned when it is not (Koriat, 1997).

Why are metacognitive judgments often wrong?
Because they are inferences from cues rather than direct read-outs of memory. The feeling of knowing, for instance, is driven by the accessibility of partial information, so plausible but incorrect fragments inflate it, and judgments of learning are driven by momentary fluency, so studying an item right before rating it makes it feel better learned than it is (Koriat, 1993; Benjamin, Bjork, & Schwartz, 1998).

What is the tip-of-the-tongue state?
It is the strong and often correct conviction that you know a word you cannot currently retrieve, frequently accompanied by partial information such as the first letter or number of syllables. It shows that retrieval is not all-or-none and that the metacognitive feeling can report on a memory even when retrieval of it fails (Brown & McNeill, 1966).

Is the Dunning–Kruger effect real?
Poor performers do tend to overestimate their performance, and the original account attributed this to a double burden, since the skills needed to do well are the skills needed to judge doing well (Kruger & Dunning, 1999). A statistical critique argues that much of the pattern follows from regression to the mean and shrinks under proper individual-differences methods, so how much reflects a specific metacognitive deficit remains debated (Gignac & Zajenkowski, 2020).

How can I make my self-assessments more accurate?
Change the procedure rather than trying harder to be accurate. Judge your learning after a delay rather than immediately, which forces the judgment to draw on durable memory, and use retrieval practice, which provides direct feedback about what you actually know and corrects the illusions that rereading creates (Nelson & Dunlosky, 1991; Dunlosky, Rawson, Marsh, Nathan, & Willingham, 2013).

Where in the brain is metacognition?
The accuracy of judgments about one's own performance is associated with anterior and rostrolateral prefrontal cortex, alongside cingulate and insular regions that track uncertainty. Individual differences in metacognitive accuracy correlate with gray matter volume in anterior prefrontal cortex, holding task performance constant (Fleming & Dolan, 2012; Fleming, Weil, Nagy, Dolan, & Rees, 2010).

References

1	Benjamin, A. S., Bjork, R. A., & Schwartz, B. L. (1998). The mismeasure of memory: When retrieval fluency is misleading as a metamnemonic index. Journal of Experimental Psychology: General, 127(1), 55–68. https://doi.org/10.1037/0096-3445.127.1.55
2	Brown, R., & McNeill, D. (1966). The "tip of the tongue" phenomenon. Journal of Verbal Learning and Verbal Behavior, 5(4), 325–337. https://doi.org/10.1016/S0022-5371(66)80040-3
3	Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & 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. https://doi.org/10.1177/1529100612453266
4	Dunning, D. (2011). The Dunning–Kruger effect: On being ignorant of one's own ignorance. Advances in Experimental Social Psychology, 44, 247–296. https://doi.org/10.1016/B978-0-12-385522-0.00005-6
5	Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry. American Psychologist, 34(10), 906–911. https://doi.org/10.1037/0003-066X.34.10.906
6	Fleming, S. M. (2024). Metacognition and confidence: A review and synthesis. Annual Review of Psychology, 75, 241–268. https://doi.org/10.1146/annurev-psych-022423-032425
7	Fleming, S. M., & Dolan, R. J. (2012). The neural basis of metacognitive ability. Philosophical Transactions of the Royal Society B: Biological Sciences, 367(1594), 1338–1349. https://doi.org/10.1098/rstb.2011.0417
8	Fleming, S. M., Weil, R. S., Nagy, Z., Dolan, R. J., & Rees, G. (2010). Relating introspective accuracy to individual differences in brain structure. Science, 329(5998), 1541–1543. https://doi.org/10.1126/science.1191883
9	Gignac, G. E., & Zajenkowski, M. (2020). The Dunning-Kruger effect is (mostly) a statistical artefact: Valid approaches to testing the hypothesis with individual differences data. Intelligence, 80, 101449. https://doi.org/10.1016/j.intell.2020.101449
10	Hart, J. T. (1965). Memory and the feeling-of-knowing experience. Journal of Educational Psychology, 56(4), 208–216. https://doi.org/10.1037/h0022263
11	Koriat, A. (1993). How do we know that we know? The accessibility model of the feeling of knowing. Psychological Review, 100(4), 609–639. https://doi.org/10.1037/0033-295X.100.4.609
12	Koriat, A. (1997). Monitoring one's own knowledge during study: A cue-utilization approach to judgments of learning. Journal of Experimental Psychology: General, 126(4), 349–370. https://doi.org/10.1037/0096-3445.126.4.349
13	Koriat, A., & Bjork, R. A. (2005). Illusions of competence in monitoring one's knowledge during study. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31(2), 187–194. https://doi.org/10.1037/0278-7393.31.2.187
14	Kruger, J., & Dunning, D. (1999). Unskilled and unaware of it: How difficulties in recognizing one's own incompetence lead to inflated self-assessments. Journal of Personality and Social Psychology, 77(6), 1121–1134. https://doi.org/10.1037/0022-3514.77.6.1121
15	Metcalfe, J. (2009). Metacognitive judgments and control of study. Current Directions in Psychological Science, 18(3), 159–163. https://doi.org/10.1111/j.1467-8721.2009.01628.x
16	Nelson, T. O., & Dunlosky, J. (1991). When people's judgments of learning (JOLs) are extremely accurate at predicting subsequent recall: The "delayed-JOL effect." Psychological Science, 2(4), 267–270. https://doi.org/10.1111/j.1467-9280.1991.tb00147.x
17	Nelson, T. O., & Narens, L. (1990). Metamemory: A theoretical framework and new findings. Psychology of Learning and Motivation, 26, 125–173. https://doi.org/10.1016/S0079-7421(08)60053-5
18	Schoenfeld, A. H. (1985). Mathematical problem solving. Academic Press.
19	Schraw, G., & Moshman, D. (1995). Metacognitive theories. Educational Psychology Review, 7(4), 351–371. https://doi.org/10.1007/BF02212307