Memory Confidence vs. Accuracy: What the Evidence Shows

Author: Angela Bogdanova · Published: October 12, 2026 · Editorial Policy

Research cluster: Memory & Forgetting · Written by Artificial Sapiens

Memory confidence is a person's subjective judgment about how likely a remembered answer, recognition decision, or account of an event is to be correct. Memory accuracy is the correspondence between that answer and independently established information about what occurred. Research shows that confidence can carry useful information about accuracy, particularly when it is reported immediately after an uncontaminated recognition decision. The relationship also depends on the type of memory task, the quality of the original experience, the conditions of retrieval, and events occurring after remembering.

The strongest conclusion is conditional: high confidence often accompanies correct memory, but confidence alone cannot establish that an event occurred as remembered. In Wixted and Wells's major synthesis, initial confident eyewitness identifications from fairly constructed, properly administered lineups predicted accuracy substantially better than older blanket claims suggested. Fitzgerald, Hyman, and Wade's 2025 field-data meta-analysis simultaneously found consequential high-confidence mistakes and emphasized suspect-related biases. A 2026 review by Yu and Moore argues that individual differences in metacognitive monitoring and bias remain central to evaluating this relationship in real situations.

What do memory confidence and memory accuracy mean?

When someone says, “I am 95% certain I saw that person,” the statement expresses confidence: a judgment about the memory. Accuracy refers to whether the identification is objectively correct, based on suitable independent evidence. One concerns subjective certainty; the other concerns correspondence to reality. Accuracy can sometimes be assessed in an experiment using known targets or recorded events. In a personal memory of an undocumented conversation, the true details may be impossible to establish, even if the individual provides a precise confidence percentage.

Confidence can attach to very different judgments: recognizing a face, recalling a name, retrieving a studied word, reporting what a friend said, locating an object, or deciding where a piece of information was encountered. Studies of one kind of judgment cannot automatically determine how confidence performs in every other kind. Research also separates confidence reported after a completed answer from predictions of future remembering, such as “I will know this tomorrow.” These involve related forms of metacognition but answer different questions. A study of face recognition found that postdecision confidence was more diagnostically informative than predictive judgments of learning for the tested faces (postdictive and predictive confidence experiments).

There is another distinction between a person's confidence on individual trials and differences between people. Within one person, higher-confidence answers may be more often correct than lower-confidence answers. Across people, someone who is generally very confident may or may not outperform someone who expresses uncertainty more freely. A meta-analysis covering 213 datasets found a modest across-person correlation between mean confidence and mean accuracy, R = .22, with meaningful variation by task domain and rating scale. This does not contradict a stronger within-task link under particular experimental conditions.

Why do people feel confident when remembering?

A confidence judgment draws on evidence available at retrieval and on the person's interpretation of that evidence. A recalled detail may arrive quickly, feel familiar, fit a coherent narrative, or seem unusually vivid. These experiences often contain useful information because stronger memory evidence can make correct answers easier to retrieve. Yet they can also arise from repetition, familiarity without the correct source, or the effortless completion of a plausible but mistaken narrative.

Memory strength, familiarity, and recollection

In a recognition task, a familiar face or word may produce a strong sense of having been encountered. Recollection can add contextual information about when, where, and how the experience occurred. These are psychologically meaningful cues, but familiarity does not identify its own origin, and even detailed recollection can contain errors. A 2012 review of neuroimaging and eye-movement findings described multiple cues used in confidence judgments, including characteristics of the retrieval prompt rather than only the original episode (Evaluating Confidence in Our Memories).

Consider seeing a person several times in a neighborhood and later encountering a police lineup. Their face might feel familiar because of ordinary prior encounters. If that familiarity is attributed to witnessing a crime, a confident identification may be mistaken even when the familiarity itself is genuine. The crucial question becomes whether the evidence supporting confidence discriminates the remembered event from other possible sources.

Fluency, coherence, and repetition

Ease of retrieval may support confidence. A detail repeated in conversation can later be retrieved fluently, and a coherent account can seem convincing when the mind fills gaps using background knowledge. Neither fluency nor coherence guarantees factual fidelity. In sentence-memory experiments, Brewer and colleagues found a stronger confidence–accuracy relation for straightforward recalled sentences than for misleading substitutions or schema-based inferences. Related experiments showed that, for deceptive recognition items involving plausible gist substitutions, high confidence could be associated with errors (metamemory approach).

This provides a practical example of why the question matters: remembering the central meaning of a discussion may be useful and approximately correct, while a person could confidently attribute exact words to a speaker who never used them.

Vividness and subjective reliving

Vivid imagery and the feeling of reliving an experience can increase subjective conviction. Sometimes vivid memories are accurate because they contain richer retrieved information. But vividness is not an independent authenticity test. Recent research testing recognition and source memory found that even vividly experienced, confident memories were not invariably factually accurate (experience of remembering study). A mentally vivid scene should therefore be evaluated using the available external evidence and how the memory was elicited, rather than the intensity of its subjective presentation alone.

What does the evidence say about the confidence–accuracy relationship?

The literature supports a relationship whose size and meaning change with the experimental design. For relatively clean recognition tests, confidence usually rises as the probability of a correct answer rises. More complex recall, misleading materials, contaminated identification procedures, and postevent feedback can weaken or distort that relationship.

In two recognition experiments using studied words and faces, Tekin and Roediger found high recognition confidence associated with high accuracy across several different rating-scale ranges. This is strong evidence against the claim that confidence is inherently unrelated to whether a response is correct. Yet recognizing a previously studied word is a different task from reconstructing a conversation under social pressure or identifying a stranger after a stressful event.

A classic false-memory demonstration reveals the other side of the evidence. Participants studied related words, such as terms associated with sleep, and later sometimes recalled or recognized the central word that was never presented. Roediger and McDermott's 1995 experiments documented confident false recognition. A 2004 quantitative review found that ratings of confidence for false recognition of nonpresented themes could resemble confidence for correctly recognized items. Laboratory paradigms of this kind establish the possibility of confident memory errors; their error frequencies should not be treated as population rates for every autobiographical memory.

Thus the central scientific result has two parts: confidence can discriminate stronger from weaker memories, and specific memory mechanisms or testing conditions can produce errors that feel convincing. Interpreting any confident account requires knowing which kind of evidence and which task generated the judgment.

How researchers measure whether confidence predicts accuracy

A simple correlation between confidence ratings and right-or-wrong outcomes is often insufficient. Three related concepts help interpret the results: calibration, discrimination or resolution, and bias.

Calibration: does stated confidence match the observed success rate?

A well-calibrated person using a probability scale would be correct on approximately 80% of the answers assigned 80% confidence, across enough comparable decisions. Calibration is assessed across groups of predictions, never confirmed from a single response. If 80%-confident responses are correct only 60% of the time, the set of judgments is overconfident. If they are correct 95% of the time, the set is underconfident. The numbers here are an illustration of the method, not estimates of ordinary human memory performance.

Calibration depends on the response categories used, how questions are sampled, and whether the study includes decisions people prefer to withhold. A 100-point scale does not create 100 levels of psychological precision. In experimental comparisons, coarse and fine confidence scales yielded comparable recognition confidence–accuracy patterns when responses were grouped appropriately (Tekin and Roediger, 2017).

Resolution: can confidence distinguish correct from incorrect answers?

Resolution concerns whether a person is more confident on correct answers than incorrect answers, even if their numbers are systematically inflated or conservative. Someone may overstate probabilities on every answer while still reliably ranking their own strong memories above weak ones. Conversely, average confidence can look well calibrated while distinguishing individual correct and incorrect judgments poorly.

The difference helps explain why one statistic cannot summarize metamemory. Confidence is useful for some decisions because it discriminates quality within a particular setting; it may still be biased when compared with absolute correctness.

Confidence–accuracy characteristic curves and eyewitness research

Eyewitness researchers often group suspect identifications into low-, medium-, and high-confidence categories and calculate accuracy for each category, producing a confidence–accuracy characteristic, or CAC, analysis. This asks a question that directly concerns real identification decisions: among suspect identifications made at a given confidence level, what proportion are correct? The results must be interpreted in light of the design, especially the target-present and target-absent lineups, the definition of a suspect identification, the proportion of guilty and innocent suspects, and the construction of the comparison group.

A separate receiver operating characteristic analysis examines how accurately a testing procedure distinguishes targets from non-targets across decision thresholds. It is used to compare discriminability, and it is not equivalent to a CAC plot. Wixted and Wells's synthesis and the National Academies assessment show why definitions, lineup procedures, and analysis choices matter when apparently competing findings are evaluated.

Why a strong association is not a certainty guarantee

A statistically strong confidence–accuracy association does not mean every high-confidence statement is true. It may describe an average trend within a sampled class of decisions. The remaining error rate depends on the environment, item difficulty, sources of bias, and the way the relevant decision was selected. It is incorrect to convert a population-level result into a categorical assurance about one eyewitness or one personal memory. A confident statement is evidence about the person's assessment, and sometimes evidence about likely accuracy; corroboration and retrieval history remain essential.

Eyewitness memory: when can initial confidence be informative?

Eyewitness identification provides an especially important test because researchers can separate correct and mistaken identifications and because the consequences of error are substantial. Historically, the relationship between eyewitness confidence and identification accuracy was often summarized as weak. Later analyses reexamined how confidence was measured and whether the identification procedures themselves introduced distortions.

In their influential 2017 review, Wixted and Wells argued that a witness's immediate confidence in an initial suspect identification can be highly informative when the identification is made under suitable, uncontaminated conditions. Their conditions include a fair lineup, an administrator who does not reveal the suspect's identity or influence the choice, and a contemporaneous confidence statement. The inference is about initial identifications, not confidence reported months later at trial.

What counts as a fair or uncontaminated identification?

A fair lineup includes fillers who match the general description sufficiently that the suspect does not stand out for irrelevant reasons. The administrator should avoid signaling who is suspected, the witness should understand that the perpetrator might be absent, and the original identification and expressed confidence should be recorded before feedback. These safeguards minimize predictable alternative explanations for a confident choice. The National Research Council's consensus report recommends blinded procedures, standardized instructions, recording confidence in the witness's own words, and video documentation.

These procedures are vital because confidence cannot repair poor encoding, biased lineup construction, or prior familiarity with an innocent suspect. They aim to make confidence more interpretable while reducing preventable errors.

What newer evidence adds to the debate

A 2025 meta-analysis by Fitzgerald, Hyman, and Wade reported that approximately one in eight high-confidence identifications in analyzed real-investigation field data administered by blind officers were known errors in which a witness selected a filler. The study also emphasized that suspect bias, social-media exposure, and misplaced prior familiarity may produce additional risks. This figure is a finding about the analyzed field data and known filler errors; it is not the universal false-identification probability for every high-confidence suspect choice or every police lineup.

The updated 2026 review by Yu and Moore frames the issue through metacognitive monitoring, judgment bias, and individual competence. It recognizes a meaningful general relationship while questioning how directly one confidence statement can be translated into an accuracy probability in real legal settings. The resulting evidence-based approach evaluates initial confidence together with the provenance of the memory, lineup quality, potential contamination, and known limitations of the statistical model.

Why courtroom confidence is a different kind of evidence

By the time testimony is presented in court, a witness may have seen photographs, discussed an identification, heard the views of investigators, or repeated the story in interviews. These experiences can alter confidence and sometimes memory content. A strong courtroom declaration can therefore differ markedly from the person's first, uncontaminated judgment. The National Academies report specifically cautions against treating courtroom confidence as a reliable substitute for documented initial confidence.

This article addresses how confidence functions as a memory judgment. For detailed questions about identification procedures, suspect selection, and investigative interviewing, the separate eyewitness-memory topic requires its own dedicated treatment.

How confidence becomes inflated after remembering

Confirming feedback and the postidentification feedback effect

Imagine identifying a face with some hesitation and then being told that investigators agree with the choice. The approving response adds new information about social authority and perceived correctness. When the person later evaluates their certainty, this feedback may become entangled with the remembered identification experience.

A meta-analysis of 20 experimental tests involving more than 2,400 participant-witnesses found robust effects of confirming feedback on later certainty and related retrospective judgments. Smalarz and Wells experimentally showed that postidentification feedback can also impair observers' ability to distinguish accurate from mistaken identification testimony.

A later study of 907 participants across two experiments found that both conventional feedback and misleading descriptions of the witness's earlier certainty affected retrospective confidence. Even some participants who received no feedback became more confident as time passed (Greenspan and Loftus, 2020). These findings explain the practical urgency of recording an initial confidence statement rather than trying to reconstruct it later.

Repeated discussion and social agreement

A memory may be repeatedly narrated among friends or family members. Agreement can make the account feel socially verified even when everyone is relying on the same imperfect initial source. Repetition may reinforce accessible wording and familiar details, while later speakers can inadvertently introduce additional information. The scientific concern is not simply that talking is bad for memory; it is that information acquired after an event may become hard to distinguish from what was originally perceived.

Studies of multiple postidentification feedback signals indicate that their effects on confidence can accumulate under tested conditions (Smalarz and Wells, 2020). The amount of change depends on the feedback and the particular task, so these findings should not be used to estimate how much any private conversation altered a person's confidence.

Source confusion and misleading information

Source monitoring refers to determining where a remembered detail originated: direct perception, someone else's report, imagination, an earlier photograph, or an inference. An individual may correctly remember a fact while incorrectly remembering its source, or may incorporate a plausible detail into a narrative about an event. The resulting conviction can seem like firsthand recollection.

Confident source mistakes and false recognitions are documented in laboratory research; their mechanisms help explain why recollection may be authentic as an experience while particular details remain uncertain (Roediger and McDermott; McKelvie's quantitative review). Source attribution is therefore part of assessing accuracy, not merely a question of how clear a memory feels.

Different memory tasks produce different confidence signals

Recognition versus free recall

Recognition asks whether a presented option was encountered before. Free recall asks someone to produce an answer without a supplied target. Recognition offers candidate information that can trigger familiarity; recall depends on different retrieval demands and may involve reconstructing details. As a result, recognition confidence findings are informative but do not automatically describe narrative accuracy. Experiments on the recall of misleading sentences demonstrate that content and reconstruction affect the value of confidence (Brewer and colleagues).

Answers, exact wording, and the gist of an event

A person can accurately remember the gist of a lecture or argument while confidently reporting an incorrect quotation. Gist can be highly useful for comprehension, even though it is insufficient evidence for exact wording. In the word-list false-memory paradigm, nonpresented but thematically related words can be falsely recognized with confidence (Roediger and McDermott, 1995). Exact wording should be checked against recordings or records when the distinction matters.

Predicting future memory versus assessing a completed answer

Judgments of learning often predict how well a person expects to remember information later; retrospective confidence follows a present memory decision. These judgments draw on partially overlapping cues, but the link to accuracy differs with task structure and timing. The face-recognition study comparing predictive and postdictive judgments found postdecision confidence more strongly related to recognition accuracy under the tested conditions. For studying, this supports using later self-testing and feedback to assess actual performance rather than relying only on how familiar material feels during reading.

Personal events and autobiographical memories

Autobiographical memories contain details about people, places, emotions, and the personal meaning of past experiences. Often there is no complete external recording with which to score the memory. Confidence may reflect a blend of retrieved detail, narrative familiarity, personal importance, and reinforcement across retellings. The absence of independent verification is an epistemic limit, not proof that a memory is incorrect. Similarly, vividness and sincerity do not establish the exactness of all details.

Can confidence be low when memory is accurate?

Yes. People differ in their willingness to express certainty, and a cautious person may provide a correct answer while assigning it a relatively low confidence rating. A momentary inability to retrieve supporting details may reduce confidence even when the selected answer is right. Metacognitive bias concerns a general tendency to use the confidence scale liberally or conservatively; metacognitive sensitivity concerns how effectively confidence distinguishes the person's own correct from incorrect decisions.

The 2022 meta-analysis by Jin, Verhaeghen, and Rahnev demonstrates why raw confidence differences between individuals cannot be treated as direct rankings of memory skill. The 2026 review likewise emphasizes competence and judgment bias when translating confidence ratings into accuracy estimates.

Low confidence can signal weak evidence, but it does not logically imply error. The appropriate interpretation asks whether the person can discriminate their own correct responses from incorrect ones across comparable tasks, and whether the scale has been calibrated against known outcomes.

Can high confidence accompany completely false memories?

Yes. Controlled experiments have produced confident recognition of words that participants had not studied. The classic Deese–Roediger–McDermott experiments and subsequent quantitative evidence establish that high subjective certainty is compatible with objectively false recognition. This does not make every strongly held autobiographical memory false, nor does it license assuming that an intense recollection is fabricated.

A useful distinction is between the subjective memory experience and the historical claim embedded in it. People can sincerely experience remembering, while an element of what they report is incorrect. Evidence about details can come from records, timelines, independent observations, and careful investigation of alternative sources. For highly consequential accounts, neither automatic acceptance nor automatic dismissal on the basis of confidence alone is scientifically justified.

Does time make confidence less reliable?

Retention interval can affect the quantity and quality of accessible memory evidence. However, the relationship between time, accuracy, and confidence is not captured by a universal rule in which confidence necessarily becomes useless after a fixed number of days. In eyewitness lineup studies, research on short and long delays reported a meaningful confidence–accuracy relationship at both intervals under studied conditions. This is evidence against a simple claim that time alone eliminates the diagnostic value of initial recognition confidence.

The time at which confidence is measured matters independently of the time between event and test. Immediate confidence after a first identification may be useful, while a statement made much later after feedback and repeated exposure can be altered. The National Academies guidance accordingly prioritizes the verbatim initial judgment. Evaluators should distinguish the event-to-test delay from the test-to-confidence delay and from subsequent social influence.

Can people improve how accurately they judge their own memories?

Calibration requires comparing confidence with outcomes rather than merely increasing how sure a person feels. For questions with verifiable answers, one can record an answer, rate the probability it is right, and later compare ratings with actual correctness across multiple trials. This can reveal whether high-confidence categories systematically contain more correct answers. The exercise measures a specific task and does not certify the reliability of all autobiographical memories.

In laboratory research, feedback can improve calibration in selected eyewitness tasks. A study of practice and feedback found changes in confidence–accuracy relationships with informative feedback, while practice without feedback was insufficient. Feedback should concern an objectively verified answer and be interpreted as a learning signal; social approval that falsely implies correctness can have the opposite effect by inflating confidence.

A practical routine for studying is to answer questions before checking notes, record uncertainty when possible, compare responses with an answer key, and revisit the kinds of errors that were incorrectly judged as certain. Such practice supplies information about retrieval performance. It cannot guarantee generalized improvements in all memory judgments, and anecdotal feelings of improved memory are distinct from measured accuracy.

How to evaluate a confident memory in everyday life

First, specify the claim. A statement about recognizing a person, remembering the gist of a conversation, or recalling exact words requires different types of verification. Clear definition prevents a broad sense of certainty from being mistakenly applied to details that were never independently checked.

Second, examine original access to the event. Was the person able to perceive the relevant detail? Were lighting, distance, distraction, and duration adequate for the specific observation? These factors concern the quality of initial encoding rather than the confidence expressed afterward.

Third, document the earliest version of the account where practical. A contemporaneous note or recording helps distinguish what the person originally reported from later additions. A national scientific assessment of eyewitness evidence recommends exactly this kind of early confidence documentation for police identification, in part because later judgments can be swayed.

Fourth, trace what was learned after the event. Did photographs, discussion, suggested interpretations, praise, or news reports introduce details that might be confused with the original experience? An account supported by independent contemporary records carries different evidential weight from one supported solely by repeated agreement within the same social group.

Fifth, seek relevant corroboration without treating every disagreement as proof of falsity. Independent records and mutually independent witnesses provide additional information. If no external record exists, distinguish what is remembered with high confidence from what can actually be established. An unresolved historical question may remain unresolved even when the subjective memory is very strong.

What confidence should mean in education, healthcare, and public decisions

In education, feeling certain while reviewing familiar material is different from answering correctly without prompts. Learners benefit from checking actual retrieval and discovering which “obvious” answers were wrong. The educational application is primarily a lesson in measurement: judge learning against performance, not against a pleasant sense of familiarity.

In healthcare, an account of when a symptom began can be important, yet uncertainty about timing or sequence is common and may warrant checking dated notes and clinical records. Memory confidence itself is not a diagnostic test for a neurological or psychiatric disorder. Clinical interpretation depends on the full history, relevant symptoms, and professional assessment rather than a person's confidence rating.

In investigations, workplaces, and family disputes, confidence can influence whether a story is believed. The evidence favors documenting initial accounts and separating observation from interpretation, particularly where social feedback may shape later recollection. A respectful approach recognizes the reality of the person's experience while evaluating specific historical claims through the best available independent evidence.

Common myths about memory confidence

Myth: “A person who is completely certain must be right.”

High confidence increases the likelihood of correctness in some well-controlled recognition tasks, but known high-confidence errors exist in false-recognition experiments and real eyewitness data. The observed relationship is probabilistic and conditional (Wixted and Wells, 2017; Fitzgerald and colleagues, 2025).

Myth: “Confidence says nothing about accuracy.”

This reverses the error. Properly measured confidence is often informative. Recognition experiments and eyewitness identification syntheses demonstrate meaningful patterns, especially for immediate, uncontaminated decisions. The challenge is to establish the relevant conditions before interpreting an individual statement (Tekin and Roediger; Wixted and Wells).

Myth: “If a memory feels vivid, the details are verified.”

Vividness concerns subjective experience, and veracity concerns external facts. A vivid memory can contain incorrect sources or details. Recent recognition and source-memory experiments directly illustrate this separation (2025/2026 report on experience of remembering).

Myth: “Repeating a story always makes it more accurate.”

Repetition can support access to an account while also stabilizing later wording and socially acquired details. Experimental feedback research shows that repeating an identification after being told it was correct can inflate confidence without supplying new evidence about what was originally seen (Douglass and Steblay; Greenspan and Loftus).

Myth: “A single confidence percentage tells us the probability that the memory is correct.”

Confidence categories become interpretable as probability estimates only when calibration has been examined in a defined class of comparable decisions. Personal confidence is not automatically a validated statistical probability. Question difficulty, response selection, base rates, scale use, and information contamination must also be considered.

Frequently asked questions

Is memory confidence correlated with memory accuracy?

Often yes, especially within controlled recognition tasks and initial, properly administered eyewitness identifications. The size and usefulness of the association differ between tasks and individuals. A large across-participant meta-analysis found only a modest mean-level correlation across people, demonstrating that general confidence should not be equated with general memory ability (Jin et al., 2022).

Are high-confidence eyewitnesses usually accurate?

High initial confidence can be a useful indicator under fair and uncontaminated testing conditions, but it does not guarantee correctness. Research based on actual investigations documents meaningful known errors among high-confidence identifications, and suspect bias remains important. Later courtroom confidence has additional risks because of intervening information and feedback (Fitzgerald et al., 2025; National Research Council, 2014).

What is the difference between confidence, vividness, and accuracy?

Confidence is a judgment about correctness; vividness is a quality of the memory experience; accuracy is correspondence to what occurred. The three can correlate, but they are different variables. A vivid, confident image can still contain a mistaken detail or source.

Can you be 100% confident and still be wrong?

Yes. Highly confident false recognition is experimentally documented. A 100% self-rating reports a psychological state, not a guarantee of historical accuracy. Roediger and McDermott's research provides clear experimental examples of confident memory errors.

Does memory confidence get stronger when an event is discussed repeatedly?

It can. Agreement, repetition, and confirming feedback sometimes increase later confidence, including when the original identification was mistaken. Whether a particular discussion increased confidence cannot be established from repetition alone (Greenspan and Loftus, 2020).

Is low confidence a sign that the memory is false?

No. Low confidence often marks uncertainty, but correct answers can be reported cautiously. Differences in personal rating style can affect the numbers independently of correctness. Calibration requires multiple independently scored judgments.

Why can someone remember the meaning of a conversation but misremember the words?

Memory may preserve gist while reconstructing precise wording. Similar or expected words can be substituted in recall and recognition, creating plausible errors. Experiments on sentence memory demonstrate that such substitutions can weaken the confidence–accuracy relationship (Brewer et al., 2005).

What does confidence calibration mean?

Calibration compares reported confidence with actual success rates across comparable decisions. If answers rated 70% confident are correct about 70% of the time, that category is approximately calibrated. A single memory is insufficient for a calibration assessment.

Is recall confidence the same as recognition confidence?

No. Free recall and recognition impose different retrieval demands and provide different cues. Confidence findings should be interpreted in the task where they were measured instead of being generalized from a face lineup or word-recognition test to every type of personal narrative.

Can someone train themselves to stop having false memories?

There is no established method that eliminates false memories in everyday life. Recording information at the time, checking independent evidence, recognizing suggestive feedback, and comparing predictions against outcomes can improve information practices. They do not make autobiographical memory infallible.

Should a court rely on a witness's confidence?

Confidence is one component of evidence. Scientific reviews recommend giving particular attention to the first confidence statement under adequately safeguarded identification procedures and examining the risk of contamination, bias, and alternative explanations. The legal assessment also requires other evidence and the applicable legal standards (National Research Council; Yu and Moore).

Related English Hub articles

For the broader psychology of memory representation and retrieval, see Human Memory: Types, Processes, and How It Works, Memory Retrieval: Recall, Recognition, and Cues, and False Memories: How They Form and What Research Shows. These published English resources provide useful context while this article retains its distinct focus on confidence–accuracy calibration.

Conclusion

Memory confidence carries meaningful information under some conditions. It can track accuracy in a clean recognition decision, yet the same subjective certainty can be influenced by gist-based reconstruction, source confusion, feedback, and social repetition. The key to evaluating confidence is to identify precisely what was remembered, when certainty was recorded, how the memory was tested, and which independent evidence supports the account. This framework respects confident human experience while evaluating factual claims with methods suited to the relevant memory task.

Related Articles

False Memories: How They Form and What Research Shows

Flashbulb Memories: Why Vivid Memories Can Be Inaccurate

Imagination Inflation: Can Imagining Events Create False Memories?

References

Brewer, W. F., Sampaio, C., & Barlow, M. R. (2005). Confidence and accuracy in the recall of deceptive and nondeceptive sentences. Journal of Memory and Language, 52(4), 618–627. https://doi.org/10.1016/j.jml.2005.01.017

Douglass, A. B., & Steblay, N. (2006). Memory distortion in eyewitnesses: A meta-analysis of the post-identification feedback effect. Applied Cognitive Psychology, 20(7), 859–869. https://doi.org/10.1002/acp.1237

Fitzgerald, R. J., Hyman, I. E., Jr., & Wade, K. A. (2025). Error rates for high confidence eyewitness identifications. Memory, 33(9), 1041–1053. https://doi.org/10.1080/09658211.2025.2546919

Greenspan, R. L., & Loftus, E. F. (2020). Eyewitness confidence malleability: Misinformation as post-identification feedback. Law and Human Behavior, 44(3), 194–208. https://doi.org/10.1037/lhb0000369

Jin, S., Verhaeghen, P., & Rahnev, D. (2022). Across-subject correlation between confidence and accuracy: A meta-analysis of the Confidence Database. Psychonomic Bulletin & Review, 29, 1405–1413. https://doi.org/10.3758/s13423-022-02063-7

McKelvie, S. J. (2004). False recognition with the Deese-Roediger-McDermott-Reid-Solso procedure: A quantitative summary. Perceptual and Motor Skills, 98, 1387–1408. https://doi.org/10.2466/pms.98.3c.1387-1408

National Research Council. (2014). Identifying the culprit: Assessing eyewitness identification. National Academies Press. https://doi.org/10.17226/18891

Perfect, T. J., Hollins, T. S., & Hunt, A. L. (2000). Practice and feedback effects on the confidence-accuracy relation in eyewitness memory. Memory, 8(4), 235–244. https://doi.org/10.1080/096582100406793

Roediger, H. L., III, & McDermott, K. B. (1995). Creating false memories: Remembering words not presented in lists. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21(4), 803–814. https://doi.org/10.1037/0278-7393.21.4.803

Smalarz, L., & Wells, G. L. (2014). Post-identification feedback to eyewitnesses impairs evaluators' abilities to discriminate between accurate and mistaken testimony. Law and Human Behavior, 38(2), 194–202. https://doi.org/10.1037/lhb0000067

Smalarz, L., & Wells, G. L. (2020). Do multiple doses of feedback have cumulative effects on eyewitness confidence? Journal of Applied Research in Memory and Cognition, 9(4), 508–518. https://doi.org/10.1016/j.jarmac.2020.06.003

Tekin, E., & Roediger, H. L., III. (2017). The range of confidence scales does not affect the relationship between confidence and accuracy in recognition memory. Cognitive Research: Principles and Implications, 2, 49. https://doi.org/10.1186/s41235-017-0086-z

Wixted, J. T., & Wells, G. L. (2017). The relationship between eyewitness confidence and identification accuracy: A new synthesis. Psychological Science in the Public Interest, 18(1), 10–65. https://doi.org/10.1177/1529100616686966

Yu, C., & Moore, K. N. (2026). Revisiting the confidence-accuracy relationship in eyewitness identification: A metacognitive perspective. Memory, 34(8), 1008–1029. https://doi.org/10.1080/09658211.2026.2691574