Metacognition is cognition about cognition: thinking about one's own thinking. The term names the construct that lets a learner not merely process information but step back and ask whether the processing is working. A reader who realises mid-paragraph that she has understood nothing, and rereads, has just exercised metacognition. As a theory it explains why two learners with equal raw ability can perform unequally, since one tracks and manages her own understanding while the other does not. The practical strategies that flow from it are inventoried separately under metacognitive strategies; this note is the underlying construct those strategies draw on.
John Flavell introduced the term in a 1979 paper that became the field's starting point. He split metacognition into two broad parts. Metacognitive knowledge is the stored, largely declarative knowledge a person has about cognition, and it has three categories. Person knowledge concerns oneself and others as thinkers ("I remember faces but not names"). Task knowledge concerns how the demands of different tasks vary ("dense academic prose needs slower reading than a news story"). Strategy knowledge concerns which approaches serve which goals ("to remember a list, group it"). Alongside this knowledge Flavell placed metacognitive experiences, the in-the-moment feelings and realisations that arise during a task, such as a sudden sense that a problem is harder than expected. His account was developmental, observing that young children hold little such knowledge and monitor their own thinking far less than adults do.
Later work crystallised a two-part structure that pairs Flavell's knowledge side with an active side. Metacognitive regulation is the set of processes that manage cognition during a task: planning before acting, monitoring while acting, and evaluating afterward. The two parts are interdependent but distinct. Knowing that rereading helps comprehension (knowledge) is useless unless the learner also notices comprehension has failed and decides to reread (regulation). This is why the distinction carries weight for teaching: a learner can possess rich metacognitive knowledge yet fail to regulate, the way someone can know the rules of a diet without following it.
Thomas Nelson and Louis Narens gave regulation a precise mechanism in their 1990 framework. They split mental activity into two levels. The object level does the basic work of encoding, learning, and remembering. The meta level holds a model of that work, of the task and the operations it requires. Two processes connect the levels asymmetrically. Monitoring sends information upward, the meta level reading off how the object level is doing. Control sends decisions downward, the meta level adjusting what the object level does, such as allocating more study time to weak items. The relationship is one-way: the meta level models the object level, but the object level has no model of the meta level. This architecture underlies most contemporary research on how people gauge and steer their own learning.
Monitoring is only useful when it is accurate, and often it is not. Calibration is the match between how well a learner thinks she has learned something and how well she actually has, measured by comparing confidence judgments (such as a judgment of learning, a prediction of future recall made during study) against later test performance. When confidence outruns competence, the result is the illusion of knowing: the learner feels she has mastered material she has not. The danger is concentrated where it does most harm. Low performers are the least calibrated, because telling knowing from not-knowing itself requires the domain knowledge they lack, so the weakest learners are the most overconfident and the least likely to keep studying. Structured judgment training with feedback can improve calibration beyond what repeated testing alone achieves.
Measuring metacognition is fraught, and the field knows it. Offline measures, taken before or after a task such as a questionnaire or interview, are convenient but suspect: self-report surveys of metacognitive monitoring frequently fail to align with actual behaviour, and are vulnerable to faking and demand characteristics. Online measures, taken during the task such as think-aloud protocols or judgments of learning, capture regulation as it happens but are themselves reactive, since asking learners to verbalise can change or even disrupt the very processing being studied, making them impractical in real classrooms. A deeper conceptual problem is that awareness is not control: a learner may accurately monitor that she does not understand yet do nothing about it, so high metacognitive knowledge does not guarantee better learning. Finally, because verbal report leans on language and on developmental maturity, it underestimates the metacognition of young children and of learners working in a weaker language.
Build both halves of the construct, not just one. Grow metacognitive knowledge by naming strategies and the tasks they fit, but pair it with regulation practice through planning prompts before a task and monitoring checks during it, since knowledge without regulation changes nothing. Attack the illusion of knowing directly: low-stakes quizzing exposes the gap between feeling-of-knowing and actual recall, and feedback on those predictions sharpens calibration over time. Watch the weakest learners most closely, as they are both the most overconfident and the least able to self-correct. Treat learner self-reports of strategy use with caution, since what learners say they do often diverges from what they do.