Polygraph Tests: How They Work, Accuracy, and Limits
Author: Ukrainian Psychological Hub · Published: October 11, 2026 · Editorial Policy
Polygraph tests measure changes in breathing, cardiovascular activity, and skin conductance while someone answers questions. The resulting recordings are interpreted within an examination protocol to form an opinion about deception or recognition of information. The instrument measures physiology directly; truthfulness is a separate psychological inference. That distinction determines what a polygraph result can mean in a criminal investigation, a government screening program, a workplace dispute, or a courtroom.
The central scientific conclusion is that some specific-incident polygraph procedures discriminate deceptive from truthful responses at rates above chance under studied conditions, but they also make consequential errors. There is no defensible universal percentage that describes the accuracy of every polygraph, every examiner, every question format, and every setting. General screening is especially uncertain. These conclusions follow the National Research Council's comprehensive evaluation and the later Iacono and Ben-Shakhar review. The American Psychological Association likewise cautions against treating physiological measurements as reliable proof of lying.
A useful way to read this topic is to keep five questions separate: What does the machine record? How does the examiner interpret the recording? How accurately does that protocol classify people in a specified population? What information does the examination add to independently collected evidence? And what, if anything, may a legal institution do with the result?
What is a polygraph test?
A polygraph examination is a structured psychophysiological assessment in which multiple body signals are recorded while an examiner asks predetermined questions. In the United States, the federal Employee Polygraph Protection Act guidance identifies cardiovascular, respiratory, and electrodermal patterns as core instrumentation. Modern systems may use computerized recording and numerical scoring, but digital equipment does not convert the recording into a direct measurement of dishonesty.
The popular term "lie detector" combines two very different things: an observable bodily response and a claim about what that response signifies. In scientific terms, the signals are measurements, a test score is an assessment output, the examiner's interpretation is an opinion, and a finding of guilt or liability belongs to the relevant legal process. Each step requires its own justification.
Polygraph testing is most familiar in investigations of particular incidents, screening for some government or security positions, and certain supervision or monitoring programs. The word "polygraph" describes a family of techniques rather than one uniform diagnostic test. Comparing studies without separating their question formats and intended uses can produce misleading claims about accuracy.
How does a polygraph work?
The physiological signals
A typical examination records respiratory changes through sensors around the torso, cardiovascular activity through a blood-pressure or related sensor, and electrodermal activity through electrodes on the skin. Electrodermal activity changes with sweat-gland activation and is linked to autonomic arousal. The channels capture different aspects of bodily regulation; none identifies lying uniquely.
These reactions can change when a question seems threatening, emotionally significant, familiar, embarrassing, ambiguous, or personally important. Attention, anticipation, uncertainty, pain, and the testing environment also matter. A truthful person under suspicion may react intensely to a question about a serious accusation. A person withholding information may show little differential response. The National Research Council's analysis of basic science treats this non-specificity as the fundamental theoretical limitation.
The examination sequence
Examinations usually include a pretest interview, a recording phase, and a posttest discussion. The examiner explains procedures, establishes the wording of questions, gathers background information, and determines which questions are relevant to the referral. Questions may be repeated in structured sequences while physiological signals are registered. An examiner or scoring procedure then evaluates response patterns against the chosen protocol's rules.
The pretest interview is part of the measurement context. It can affect expectations, anxiety, understanding of a question, and the subjective importance of its content. Consequently, examiner behavior and wording can influence the very signals that the test is supposed to interpret. Standardization and documentation are central to evaluating any claimed accuracy.
During the posttest phase an examiner may discuss charts, seek explanations for responses, or invite further statements. Admissions obtained through conversation are distinct from the physiological score. Their evidentiary value depends on the content of the statements, how they were elicited, whether they were independently corroborated, and the governing law.
Why a physiological reaction is not a lie
Consider someone who is asked whether they stole money. Increased arousal could reflect fear of wrongful accusation, awareness of incriminating facts, anxiety about employment, a traumatic memory, or deliberate concealment. The same measurement pattern has several plausible explanations. An inference to deception requires evidence that separates these explanations rather than assuming only one can produce the result.
The converse is equally important. Small physiological changes do not verify that a statement is true. A person's internal state, the relative salience of comparison questions, the measurement conditions, and the scoring threshold can all produce a reassuring-looking result without resolving the underlying event.
Which types of polygraph tests are used?
Comparison Question Test (CQT)
The comparison question technique compares physiological responses to issue-relevant questions with responses to broader comparison questions. A test might place a question about a particular alleged act alongside a question concerning past dishonesty or another personally salient topic. Interpretation assumes that the relative pattern of responses distinguishes the examinee who is deceptive about the target incident from the truthful examinee.
This assumption is contested because both truthful and deceptive examinees may react strongly to relevant questions, and the psychological meaning of comparison questions is difficult to standardize. The 2019 review in Law and Human Behavior concluded that major concerns identified in 2003 remained unresolved, including research quality and uncertainty about field error rates. Some controlled studies nevertheless find above-chance discrimination, and the National Research Council recognized that limited empirical capability.
An examiner's judgment that the charts indicate deception is therefore a protocol-dependent forensic opinion. It is not a direct observation of a lie, a psychiatric diagnosis, or a legal conclusion about guilt.
Concealed Information Test (CIT), also called the Guilty Knowledge Test
The concealed information test examines recognition of information that an investigator believes is known to a person involved in an event but not generally known to others. Several plausible alternatives may be presented, and the test evaluates whether the true event detail produces a distinctive response. The target construct is recognition or possession of information, rather than lying itself.
That narrower target has substantial experimental support. A meta-analysis by Ben-Shakhar and Elaad synthesized 80 laboratory studies and identified meaningful discrimination, with considerable variation across conditions. A review of field applications in Japan documents use in criminal investigations while discussing limits on probative value and implementation.
Recognition still does not establish participation in a crime. An innocent witness, first responder, journalist, relative, or person exposed to leaked case details might know the same information. Conversely, an involved person may not remember or recognize the selected item. Information leakage, investigators' choice of alternatives, and unequal familiarity with the answer options are central validity threats.
Directed-lie and other question formats
Some comparison-based protocols instruct examinees to give an explicitly false answer to a comparison question. Such designs seek greater standardization of the comparison condition, but a standardized question does not solve the broader problem of uniquely relating physiological reactions to deliberate deception. Studies of one protocol must not be used to declare all other protocols equally accurate.
Other technologies marketed for deception detection include voice-stress analysis, eye-movement measures, and algorithmic behavioral analysis. Their evidence must be assessed separately. The existence of a sophisticated sensor or artificial-intelligence classification algorithm does not establish clinical, investigative, or legal validity.
How accurate are polygraph tests?
Why one accuracy percentage is misleading
A statement such as "the polygraph is 90% accurate" has no adequate meaning without specifying the test format, target behavior, sample, truth-verification procedure, decision threshold, and handling of inconclusive results. Studies variously report overall classification accuracy, sensitivity, specificity, area under the receiver-operating-characteristic curve, or proportions after inconclusive tests have been removed. These metrics answer different questions.
The National Research Council found above-chance discrimination in specific-incident research involving people untrained in countermeasures, along with unstable estimates and insufficient evidence to assign a precise accuracy rate for actual practice. It also concluded that field studies were vulnerable to biases that could inflate estimated performance. The 2019 scientific update found that the foundational reliability objections remained important.
Polygraph industry figures, results from motivated volunteers, and results from a particular laboratory task cannot simply be averaged into a universally valid number. The empirical disagreement is partly about the quality and independence of the ground-truth evidence, not just about which percentage sounds most plausible.
Sensitivity, specificity, false positives, and false negatives
Sensitivity refers to the proportion of truly deceptive cases that a protocol labels deceptive. Specificity refers to the proportion of genuinely truthful cases that it labels truthful. A false positive occurs when a truthful examinee is flagged; a false negative occurs when a deceptive examinee is cleared. These terms require an independently established reference condition for truthfulness.
An "inconclusive" classification is another outcome, and excluding it from a reported accuracy rate can make a test appear stronger than its full decision process actually is. An inconclusive finding should be reported transparently, with the relevant denominator and rules for retesting.
For illustration only, imagine a hypothetical test with 80% sensitivity and 90% specificity, applied to 1,000 people among whom 20 are actually withholding the specified information. The test would correctly flag 16 of the 20 and incorrectly flag 98 of the 980 truthful people. Of the 114 positive results, only 16 would be true positives. This is arithmetic about base rates, not a validated estimate for any existing polygraph procedure.
The implication is general: even a test that performs reasonably in a balanced experiment can generate mostly false alarms when the event it seeks is rare. The National Research Council's analysis of diagnostic decisions and its security-screening chapter make this problem explicit.
Specific-incident investigation versus broad screening
A specific-incident examination asks about an identified event. In contrast, broad employee or security screening may ask whether an examinee has engaged in vaguely defined misconduct over an extended period. The examinee and examiner may interpret these questions differently, and investigators may lack an independently verifiable event against which to test accuracy.
The National Research Council concluded that the evidence supporting specific-incident examinations could not be generalized to screening and that screening accuracy was likely lower and more uncertain. Especially in low-prevalence settings, the number of nondeceptive people wrongly flagged can be much larger than the number of actual threats detected.
Screening can still influence behavior by generating anxiety, encouraging disclosures, or deterring applicants, but those administrative effects are different constructs from detecting actual deception. A program's institutional utility cannot be inferred merely from the operator's belief in its diagnostic accuracy.
Calibration and the probability of deception
A numerical score may help rank response patterns without being calibrated as an individual probability. A calibrated probability statement would require, among other things, representative validation data showing how often a specified outcome occurs among people assigned that probability. Many polygraph outcomes are categorical judgments or scores rather than validated case-level risk probabilities.
A polygraph label therefore cannot be translated automatically into "there is an 85% chance this person lied." The distinction between discrimination, calibration, and an individual's legal case is essential, particularly when a decision may affect liberty or employment.
What makes polygraph results unreliable?
Stress, fear, attention, and meaning
The physiology recorded by a polygraph is sensitive to the psychological significance of questions. The relevance of a question for an innocent person can be greater than expected, especially when the consequences of an accusation are severe. Differences in sleep, fatigue, discomfort, attention, and the examination setting can add further variation. The National Academies review identifies numerous uncontrolled sources of variability.
Health, disability, medication, and accessibility
Medical conditions, medications, and differences in autonomic response can change the interpretability of physiological recordings. The resulting concern is about suitability of the procedure and fairness in a specific context; it does not justify assuming that people with a diagnosis are more deceptive. Research does not establish robust, invariant test accuracy across all clinical, demographic, and disability groups.
A responsible assessment asks whether the relevant procedure has adequate validation for the examinee and setting, whether accommodations are possible, and whether the response pattern can be interpreted at all. No diagnosis or medication profile supplies a simple correction factor that turns a chart into a truth verdict.
Examiner influence and procedural quality
An examiner selects or approves questions, manages the testing environment, communicates expectations, and may have access to investigative claims. Lack of blinding to the suspected outcome creates a risk of confirmation bias. Differences in scoring and treatment of inconclusive cases further limit comparisons between laboratories or agencies.
The strongest validation design would independently establish truth, blind interpreters to case expectations, predefine thresholds, describe missing and inconclusive results, and evaluate the protocol on representative real-world cases. The historical evidence base rarely meets all of these conditions simultaneously, according to the National Research Council.
Deliberate countermeasures
Conscious behavioral or cognitive efforts may influence physiological signals. The National Research Council found sufficient concern that countermeasures could undermine security screening and noted uncertainty about detecting such attempts reliably. This is a limitation of the measurement system, not proof that every suspicious chart results from manipulation.
The practical scientific issue is whether validity demonstrated under controlled research conditions persists when motivated examinees differ from research volunteers. A study that excludes or inadequately represents such conditions should not be presented as measuring operational accuracy.
Cultural and language fairness
A question's meaning, perceived threat, expectations about authority, and comfort with the examiner can differ across cultural and linguistic contexts. Translation also changes the precision and emotional impact of question wording. A claim of fair classification requires actual subgroup validation and transparent error-rate reporting, rather than the assumption that identical sensors yield identical validity for every group.
Published evidence does not provide a general guarantee that polygraph error rates are stable across all sociodemographic categories. This gap is especially consequential when a screening result affects access to public employment or when a forensic opinion enters a legal proceeding.
Does passing a polygraph prove innocence?
Passing a polygraph establishes only that, under a particular protocol and threshold, the recorded responses did not meet the criteria for a deceptive classification. It cannot rule out every relevant act or demonstrate that all statements were accurate. A result described as "no deception indicated" remains an interpretation of physiological data, not independent evidence that the alleged event did not occur.
Similarly, failing a polygraph does not prove guilt. An adverse classification can occur in a truthful examinee, and even a recognition-based test can be positive because crime-relevant information was learned innocently. The legal and factual questions require corroborating records, reliable testimony, and evidence evaluated under the applicable rules.
The sensible evidentiary question is what information the examination adds beyond what investigators already know. The National Research Council found essentially no evidence establishing incremental validity over alternative methods. Treating an uncorroborated chart as decisive reverses the proper relationship between investigation and proof.
Why are polygraphs still used?
Polygraph testing persists in some investigations and security programs because it can structure an interview, create pressure to disclose information, and influence organizational decisions. These outcomes are real administrative possibilities, but they must be measured separately from accuracy at detecting lies.
The U.S. Department of Justice's archived guidance on polygraphs explicitly distinguishes an instrument recording physiological reactions from an actual lie detector. It also describes the Department's limited investigative uses. This is an official historical document; current operating policy should be checked for the particular agency and purpose.
The U.S. Customs and Border Protection applicant guidance published in 2026 states that some applicants may be required to undergo a standardized polygraph as part of the hiring process. The existence of such a program establishes the agency's administrative practice, not a scientific proof of diagnostic infallibility.
An admission made in an examination interview may lead investigators to independently check a fact. Whether it is accurate, voluntary, lawful, and independently corroborated is a separate question from whether the chart was scored as deception. High-stakes interview contexts require careful records and procedural safeguards.
Are polygraph results admissible in U.S. courts?
There is no single rule making polygraph results automatically admissible or automatically excluded from every American legal proceeding. Admissibility depends on jurisdiction, court system, specific procedural rules, applicable precedents, purpose, and the evidence offered. An examiner's opinion, a party's statement, and a recorded physiological chart may raise different evidentiary questions.
In United States v. Scheffer, 523 U.S. 303 (1998), the U.S. Supreme Court upheld the constitutionality of the military rule excluding polygraph evidence in the case before it. The decision did not establish a universal federal rule that governs every civil or criminal court and every possible use of a polygraph-related statement.
Where an expert opinion is offered in U.S. federal court, evidentiary reliability and fit are relevant under the applicable rules, including Federal Rule of Evidence 702 and its 2023 amendment. Those requirements govern admissibility of expert testimony generally; they do not certify a polygraph technique or dictate a single outcome in every case.
Rules vary by state and country, and the law may change. In a live dispute, an attorney familiar with the jurisdiction and the specific proceeding must assess admissibility, consent, discovery, and the treatment of any statement obtained during an examination. An evaluation of scientific validity is distinct from a court's legal finding.
Can an employer require a polygraph test?
Under the federal Employee Polygraph Protection Act (EPPA), most U.S. private employers are prohibited from requiring or requesting lie-detector tests for hiring or employment and from penalizing employees or applicants for exercising protected rights. There are limited exemptions, including specified security and pharmaceutical occupations and certain investigations of workplace economic loss under defined conditions.
The statute excludes federal, state, and local government employers from its private-sector coverage. Some government screening programs operate under their own legal authorities and procedures. Even when an exemption applies, the federal Department of Labor's official fact sheet sets out restrictions, written-notice requirements, standards for examiners, and protections for examinees.
An employer's use of newer instruments does not necessarily evade the law. A 2024 Department of Labor field assistance bulletin explains that certain AI-based systems used to infer dishonesty from eye, voice, or other behavior can fall within the EPPA's definition of a lie detector. Whether the statute applies depends on the actual use, employer, and legal exemption.
What should a person know before a polygraph examination?
A person asked to undergo a polygraph can reasonably request an explanation of the purpose, whether participation is voluntary or required under a particular authority, who will receive the results, how inconclusive outcomes are treated, and whether statements made during the process will be retained or disclosed. These are questions about informed participation and institutional accountability.
The exact rights depend on the setting. Employment under the EPPA, criminal questioning, agency hiring, and court-ordered supervision are governed by different legal frameworks. A person facing potential criminal or serious employment consequences should obtain appropriate jurisdiction-specific legal advice rather than interpreting a web article as individualized legal guidance.
An examiner should define the intended test protocol and its limits, distinguish recognition of information from deception, disclose material sources of uncertainty, and avoid reporting an absolute finding of truthfulness. The subject should understand that the procedure collects physiological responses while an interview can collect statements with independent consequences.
Can a polygraph identify a false confession or determine credibility?
A polygraph cannot retrospectively validate a confession merely because an examiner believes the confession matches a physiological response. Confession reliability depends on independent corroboration, contamination, questioning conditions, and alternative accounts of how case information was learned. The chart does not substitute for those investigations.
Likewise, a witness's credibility encompasses perception, memory, opportunity to observe, motivation, and consistency with external evidence. A physiological response during selected questions is too narrow and too indirect to measure all these dimensions. Polygraph findings must not be treated as a substitute for the evaluation of evidence by investigators or fact finders.
Polygraphs compared with other deception-detection approaches
Human judgments from facial expressions, eye contact, nervousness, or posture are often promoted as reliable indicators of lying. Their limited specificity shares a general problem with polygraph interpretation: a behavior can have multiple causes. The mechanistic study of nonverbal cues is a related but distinct intent that calls for its own evidence base.
Recognition-based psychophysiological testing has a narrower construct than global judgments of honesty. Its scientific strengths depend on carefully chosen undisclosed details and independent validation. The CIT meta-analysis supports the relevance of this distinction, while field-use research describes practical challenges including information leakage and how investigators evaluate knowledge.
Computer scoring may reduce one source of interpreter variability, but the result can only be as valid as the physiological measures, target construct, research population, question design, and decision threshold. Automation does not repair a conceptually invalid inference.
Common myths about polygraph tests
Myth: The machine detects lies directly
The machine records bodily signals. Deception is an interpretation of those signals under a protocol. The American Psychological Association identifies this inferential leap as the central problem.
Myth: A confident examiner can make the test error-free
Skill and protocol adherence may affect administration quality, but they cannot erase false positives, false negatives, uncertain ground truth, or the non-specificity of arousal. Examiner confidence is not an independently validated accuracy metric.
Myth: Anyone who fails must be guilty
A failed result is an assessment classification. Criminal guilt is a legal finding determined through a different process with different standards of proof.
Myth: Innocent people always pass
Fear and situational stress can occur in truthful people. The nature of comparison questions and the decision threshold may lead to adverse classifications. A normal emotional reaction to a serious accusation carries no unique moral or legal meaning.
Myth: A concealed information test proves who committed a crime
Recognition can come from many sources. The CIT tests information familiarity under specified conditions and cannot, on its own, identify the path by which a person acquired the information.
Myth: A fixed accuracy rate applies to all settings
Specific-incident investigations and broad screening differ in questions, base rates, verification possibilities, and consequences. Applying one percentage to all settings ignores the design of the test and the quality of the evidence.
Frequently asked questions
What does a polygraph measure?
It records changes in respiration, cardiovascular activity, and electrodermal activity. These are physiological responses affected by many processes, including arousal and attention.
How long does a polygraph test take?
Duration depends on the program and circumstances. For example, CBP's applicant guidance describes its own standardized hiring examination process. Its timing and procedures are not a universal standard for every polygraph.
Can anxiety cause someone to fail a polygraph?
Anxiety can influence recorded physiology, while classification depends on the relative response pattern, questions, and scoring rules. The effect on an individual's final result cannot be predicted from an anxiety label alone.
Can a polygraph be wrong?
Yes. False-positive, false-negative, and inconclusive outcomes are all possible. The rates vary with protocol, setting, threshold, and study design.
Is a polygraph the same as a medical or mental-health test?
No. It is a psychophysiological procedure used to interpret responses in an investigative or administrative context, and it does not establish a clinical diagnosis.
Can a person refuse a polygraph?
Rights and consequences depend on jurisdiction and setting. Most U.S. private-sector workers and applicants have substantial protections under the EPPA, while government employment and other contexts have distinct rules.
Are polygraphs allowed in criminal investigations?
Their use as an investigative procedure and the admissibility of their results in court are separate legal questions. Agencies and courts apply context-specific rules.
Is the concealed information test better than the comparison question test?
The CIT measures recognition of particular information and has a substantial laboratory literature; the CQT seeks an inference about deception from differential responses to categories of questions. They address different constructs, and field performance depends on suitable questions, information security, and external validation.
Do AI lie detectors solve the polygraph problem?
No automatic inference follows from using AI. Validation still requires representative ground truth, disclosed error rates, fairness analysis, and legal compliance. Some employment-oriented AI lie-detection tools can fall under existing U.S. Department of Labor guidance.
What the evidence supports
The present evidence supports a bounded conclusion. Certain polygraph protocols can distinguish groups under specified research conditions better than chance. The physiological measures remain non-specific, and operational error rates, screening validity, calibration, procedural variation, and individual interpretation remain serious limitations. The strongest empirical case for recognizing hidden information does not make recognition synonymous with guilt.
The practical standard is to describe which construct was measured, the exact method used, the target population, the independently verified outcomes, and the plausible alternatives. Polygraph results should be contextualized as uncertain assessment information whose role in an investigation or proceeding is governed by evidence, procedural fairness, and law.
For broader context, see Investigative Psychology: What It Is and How It Differs From Criminal Profiling.
Related Articles
References
American Psychological Association. (2025). Do “lie detectors” work? What psychological science says about polygraphs. APA official topic page.
Ben-Shakhar, G., & Elaad, E. (2003). The validity of psychophysiological detection of information with the Guilty Knowledge Test: A meta-analytic review. Journal of Applied Psychology, 88(1), 131–151. https://doi.org/10.1037/0021-9010.88.1.131.
Gamer, M., & Ambach, W. (2014). Deception research today. Frontiers in Psychology, 5, 256. https://doi.org/10.3389/fpsyg.2014.00256.
Iacono, W. G., & Ben-Shakhar, G. (2019). Current status of forensic lie detection with the comparison question technique: An update of the 2003 National Academy of Sciences report on polygraph testing. Law and Human Behavior, 43(1), 86–98. https://doi.org/10.1037/lhb0000307.
Matsuda, I., Nittono, H., & Allen, J. J. B. (2012). The current and future status of the concealed information test for field use. Frontiers in Psychology, 3, 532. https://doi.org/10.3389/fpsyg.2012.00532
Matsuda, I., Ogawa, T., & Tsuneoka, M. (2019). Broadening the use of the concealed information test in the field. Frontiers in Psychiatry, 10, 24. https://doi.org/10.3389/fpsyt.2019.00024.
National Research Council. (2003). The polygraph and lie detection. The National Academies Press. https://doi.org/10.17226/10420.
Osugi, A. (2018). Field findings from the concealed information test in Japan. In Detecting Concealed Information and Deception: Recent Developments (pp. 97–121). Academic Press. https://doi.org/10.1016/B978-0-12-812729-2.00005-7.
U.S. Customs and Border Protection. (n.d.). Working for CBP: Pre-employment steps and application status. CBP official applicant guidance.
U.S. Department of Justice. (Archived). Justice Manual: Polygraphs—in general, §259. DOJ official archive.
U.S. Department of Labor. (2008). Fact Sheet #36: Employee Polygraph Protection Act of 1988. Department of Labor official fact sheet.
U.S. Department of Labor. (2024). Field Assistance Bulletin 2024-1. Official PDF.
U.S. District Court, Middle District of Alabama. (2025). Amendments to Federal Rules effective December 1, 2023. Official court notice.
United States v. Scheffer, 523 U.S. 303 (1998). United States Reports, Government Publishing Office.