Picture the scene, because you have seen it a hundred times. A suspect sits in a bare room, wires on the chest, a cuff on the arm, metal plates on the fingers. A calm examiner asks a string of questions. Needles scratch across a rolling chart, or lines spike across a laptop screen. And then, the dramatic pause: the machine, we are told, knows whether the person is lying.

It is one of the most durable images in crime television, and one of the most successful pieces of marketing in the history of forensic science. The polygraph is routinely called a “lie detector,” and millions of people — job applicants, criminal suspects, security-clearance holders, television audiences — treat its verdict as something close to a verdict on truth itself.

The machine deserves a closer look. After a century of research, the honest summary is awkward for everyone: polygraphs do better than a coin flip, sometimes much better, but they do not detect lies. They detect the body’s stress responses, and stress is a noisy, promiscuous signal. Innocent people under pressure produce it. Practiced liars suppress it. And the gap between what the test actually measures and what its name promises is exactly where wrongful accusations — and escaped spies — live.

What the machine actually measures

A modern polygraph is not one sensor but a bundle of them. A blood-pressure cuff tracks cardiovascular activity. Elastic tubes around the chest and abdomen record breathing rate and depth. Electrodes on the fingers measure skin conductance — how much you sweat, which changes the skin’s electrical resistance. Some setups add a motion sensor in the seat pad, to catch a subject who tries to game the test by tensing a muscle at strategic moments.

None of these channels measures deception. There is no physiological signature unique to lying, no “lie response” that the body produces only when it utters a falsehood. What the sensors record is arousal: the fight-or-flight machinery kicking in. Heartbeat quickens, breathing shallows, palms dampen. That happens when you lie, sometimes. It also happens when you are frightened, angry, embarrassed, in pain, or simply aware that a machine is judging your honesty and your future may depend on the result.

This is the central problem, and it was understood early. The test does not read your mind. It reads your nervous system, and then an examiner infers what your nervous system meant.

How a test is supposed to work

The most common format is the Control Question Test. The examiner asks three kinds of questions. Irrelevant questions establish a baseline (“Is your name John?”). Relevant questions address the matter under investigation (“Did you take the money from the safe?”). Control questions address broad, mildly embarrassing topics that most people would answer with some discomfort (“Have you ever lied to someone who trusted you?”).

The theory is that a truthful person will react more strongly to the control questions — because everyone has something to hide there — while a deceptive person will react more strongly to the relevant questions, because those are the ones that matter. The examiner compares the size of the reactions and scores the difference.

Notice how much of this rests on psychology, not physiology. The test assumes the examiner can craft control questions that reliably provoke the right amount of discomfort in the right people, that the subject understands the questions the same way the examiner does, and that nothing else — a medical condition, a medication, a panic attack, a bad night’s sleep — distorts the signals. Each of those assumptions is a place where the test can quietly go wrong.

There are other formats, including “directed lie” tests and concealed-information tests, which look for recognition of crime details rather than deception as such. Those variants have their own strengths and weaknesses, but the Control Question Test remains the workhorse of the field, and it is the format most of the accuracy research concerns.

A century of ambition, from the lab to the courtroom

The polygraph’s origin story is older than most people realize. In the 1910s, Harvard psychologist William Moulton Marston developed an early deception test based on systolic blood pressure — the same Marston who later co-created the comic-book hero Wonder Woman, complete with a magic lasso that compelled truth-telling. The fictional lasso and the real machine were born from the same wish: a shortcut to certainty about what other people are thinking.

In 1921, John Augustus Larson, a Berkeley police officer and physiology student, built the first instrument that recorded blood pressure, pulse, and respiration together. Leonarde Keeler refined it in the following decade, adding the galvanic skin response channel, and began marketing the device to police departments. The machine spread fast, propelled by exactly the confidence its name implied.

The courts were skeptical much earlier than the public. In 1923, in Frye v. United States, a federal appeals court refused to admit Marston’s blood-pressure test as evidence, ruling that a scientific technique had to be generally accepted by the relevant scientific community before a jury could hear about it. The “Frye standard” governed the admissibility of novel science in American courts for decades, and polygraph evidence mostly failed it. That skepticism never fully went away.

What the biggest reviews actually found

If you want the most authoritative answer to the question “do polygraphs work,” the starting point is the 2003 report by the National Research Council, part of the U.S. National Academies of Sciences. A 14-member committee reviewed 57 studies encompassing 3,681 examinations and produced the most comprehensive government evaluation ever conducted. Its conclusion is quoted so often because it is so carefully balanced: specific-incident polygraph tests “can discriminate lying from truth telling at rates well above chance, though well below perfection.”

That sentence contains both halves of the story. “Well above chance” is real. In controlled laboratory studies, where volunteers are assigned to lie or tell the truth about a mock crime, polygraphs distinguish the two groups at rates far better than guessing. The committee’s analysis put median accuracy for the Control Question Test at around 85 percent in detecting deception. A later meta-analysis by the American Polygraph Association, published in 2011, covering 38 studies and 11,737 scored results, found 89 percent accuracy for single-issue diagnostic testing and 87 percent overall across validated techniques. A 2021 independent meta-analysis of 138 Control Question Test datasets, published in Applied Cognitive Psychology, again confirmed accuracy significantly above chance, with no sign of publication bias.

But “well below perfection” is doing just as much work. An accuracy rate in the mid-to-high 80s means that roughly one decision in seven or eight is wrong. And laboratory accuracy, the committee warned, is “almost certainly higher” than real-world accuracy, because real examinations involve far more variation — in examiners, in subjects, in stakes, in the questions asked — than a controlled experiment can reproduce. The committee’s blunter judgment still stands as the definitive scientific verdict: “almost a century of research in scientific psychology and physiology provides little basis for the expectation that a polygraph test could have extremely high accuracy.”

*A stethoscope on a medical desk. Like a polygraph, it measures the body — and like a polygraph, it cannot by itself tell you why the body is reacting.*

The base-rate trap: why screening is the weakest use

The polygraph’s numbers get much worse in the setting where governments use it most: screening large groups of people for rare misconduct. This is a matter of arithmetic, not opinion.

Imagine screening 10,000 employees for espionage, and suppose 10 of them are actually spies. Even a test that is 90 percent accurate in both directions — catching 9 of the 10 spies, while wrongly flagging 10 percent of the 9,990 innocent people — would produce 999 false alarms alongside 9 true catches. More than 99 percent of the people flagged would be innocent. The rarer the thing you are hunting, the more the false positives swamp the true ones.

The National Research Council saw this clearly and said so: polygraph accuracy, it concluded, “is insufficient to justify reliance on its use in employee security screening in federal agencies.” Federal agencies kept using it anyway, and the committee also warned about the human costs of that choice — innocent employees wrongly implicated, morale damaged, and talented people deterred from working for organizations that subject them to the test.

This is why the distinction between “specific-incident” testing and screening matters so much. Asking “did this person commit this particular act” is a fundamentally easier problem than asking “is this person, among thousands, hiding something.” The polygraph’s respectable laboratory numbers belong mostly to the first problem. Its reputation is used for the second.

The spies who passed

No case illustrates the gap better than Aldrich Ames. A CIA officer who sold American secrets to the Soviet Union and then Russia from the mid-1980s until his arrest in 1994, Ames passed routine counterintelligence polygraph examinations more than once while actively betraying his country — a betrayal that exposed dozens of American assets, at least ten of whom were executed. He was not an anomaly. Other spies publicly known to have beaten the polygraph include Karl Koecher, Larry Wu-tai Chin, and Ana Belén Montes, each of whom passed examinations while working for foreign intelligence services.

Ames himself was withering about the instrument afterward. In a letter to a researcher, he described its most obvious use as “a coercive aid to interrogators, lying somewhere on the scale between the rubber truncheon and the diploma on the wall behind the interrogator’s desk.” It is a vivid line, and a self-serving one — Ames had every reason to talk down the machine that failed to catch him — but the record itself is not in dispute. The machine that was supposed to protect the nation’s secrets waved through one of its worst traitors.

How did he do it? Countermeasures are the polygraph’s open secret. Because the test compares reactions to different questions, anything that artificially inflates the reaction to control questions — or dampens the reaction to relevant ones — skews the comparison. Known techniques include pressing toes hard against the floor, biting the tongue, or doing mental arithmetic, like counting backward by sevens, during control questions, so that those questions produce the bigger spike. Retired University of Minnesota psychologist David Lykken, one of the field’s most prominent critics, argued that college students given a brief explanation could beat the test, and that any working spy would have been trained to do the same.

Examiners push back, and their objection deserves a fair hearing. Modern examiners are trained to spot countermeasures, seat sensors catch some physical tricks, and the research literature includes studies in which trained examiners detected many attempts. Some analysts also argue, looking back at Ames’s charts, that signs of deception were present and the examiners simply misread them — that the failure was human, not mechanical. That defense, though, cuts both ways: a test whose accuracy depends so heavily on the individual examiner’s skill, judgment, and alertness is not the objective instrument its name suggests. It is a judgment call wearing a machine’s costume.

*An anatomical brain model. Decades of neuroscience have found no brain signal unique to lying — the reason no sensor, however sensitive, can serve as a lie detector on its own.*

Why the innocent fail

The mirror image of the trained spy is the truthful person who fails. Anxiety is the most obvious reason: a person who is terrified of being falsely accused will often show the strongest reactions to exactly the questions they most fear — the relevant ones. The test interprets that fear as deception. People with anxiety disorders, heart conditions, respiratory problems, or those taking certain medications can produce distorted readings for reasons that have nothing to do with honesty. A subject who misunderstands a question, or who is harboring an unrelated guilty secret that a control question happens to touch, can skew the comparison in the other direction.

There is also a quieter, more troubling mechanism: the examination itself is designed to be intimidating. The pre-test interview often involves the examiner asserting that the machine is essentially infallible, that deceptive answers will be obvious, and that the subject’s best course is to confess now. Subjects who believe that may confess to things they did not do, or volunteer damaging admissions under pressure. False confessions are not a hypothetical risk of high-pressure interrogation; they are a documented one, and the polygraph’s aura of scientific certainty is part of the pressure.

None of this means every polygraph result is meaningless. A skilled examiner, using validated techniques on a specific incident, with a cooperative subject, will outperform chance by a wide margin. But “better than chance” is a long way from “proof,” and it is a long way from what happens in popular culture, where a failed polygraph is treated as a confession and a passed one as an exoneration.

What the law says — and why it matters

The legal system’s verdict on the polygraph is more consistent than the scientific debate, and it is skeptical. Most federal courts and many state courts exclude polygraph evidence, citing high error rates and the risk that juries will give the results more weight than they deserve. The U.S. Supreme Court upheld the military’s ban on polygraph evidence in United States v. Scheffer in 1998, noting that the scientific community remained deeply divided on the technique’s reliability. Military Rule of Evidence 707 goes further than most civilian rules, barring not only results but even any reference to an offer to take, or refusal to take, a polygraph examination.

In employment, Congress acted even earlier. The Employee Polygraph Protection Act of 1988 prohibits most private employers from using polygraph tests for pre-employment screening, with exemptions for certain security-sensitive roles, government contractors, and specific situations such as an ongoing investigation into theft. The law exists because, before it, workers were routinely hired or fired on the basis of a test that the scientific evidence could not support at that weight.

Some jurisdictions do allow polygraph results when both sides stipulate to them in advance — a rule that, tellingly, lets the parties gamble on the machine only when they have agreed to be bound by its answer. That is not how courts treat reliable evidence. It is how they treat a coin flip both sides have agreed to respect.

*A research setting. The strongest polygraph studies are laboratory experiments — and even their authors warn that real-world accuracy is almost certainly lower.*

So why is it still everywhere?

If the evidence is this mixed and the courts are this wary, the polygraph’s survival needs an explanation. Part of it is institutional inertia: intelligence agencies, police departments, and parole systems built procedures around the test decades ago, and bureaucracies rarely abandon a tool that gives them answers, even imperfect ones. Part of it is that the test does something useful that has nothing to do with detecting lies — it makes people talk.

Researchers and even critics acknowledge this. The polygraph room, with its wires and its aura of omniscience, is a powerful interrogation prop. Subjects who believe the machine will catch them sometimes confess before the test even begins, or admit to things during the pre-test interview that investigators would never otherwise have learned. As one longtime analyst of the field put it, the polygraph functions as a prop more than anything else — yet there are cases every year in which the prop works. A tool that produces confessions is valuable to investigators even if it is a poor detector of deception, and that value helps explain why it endures.

There is also the deterrent argument: agencies contend that requiring polygraphs discourages applicants with something to hide from applying at all. That may be true. But deterrence is not detection, and it is not evidence. A scarecrow also deters crows; nobody claims it counts them.

Finally, there is the simple power of the name. “Lie detector” is one of the great brand names of the twentieth century. It promises what every interrogator, employer, and suspicious spouse wants: a direct line to the truth. The machine’s actual output — a set of squiggly lines measuring sweat and heartbeat — is far less marketable, so the brand persists and the nuance gets lost.

The Verdict

The claim that lie detector tests can reliably tell whether someone is lying is Misleading.

The evidence supports a much narrower statement: under controlled conditions, for a specific incident, administered by a trained examiner using validated techniques, a polygraph can distinguish truth-tellers from liars at rates well above chance — roughly in the mid-to-high 80 percent range in laboratory studies, with real-world performance almost certainly lower. That is genuinely better than guessing, and it is why the test has not been discarded outright.

But the popular claim goes far beyond that. A machine that is wrong roughly one time in seven, that mistakes anxiety for guilt, that can be beaten by toe-curling and mental arithmetic, that waved through Aldrich Ames and other foreign agents, and that the nation’s courts mostly refuse to admit as evidence, is not a lie detector in any sense that should settle an argument, decide a hiring decision, or brand a person a liar. It measures arousal and calls it deception, and the difference between those two things is the whole story.

The next time you see the wires and the chart on television — or the next time someone suggests that a real person’s fate should hinge on one — remember what the National Research Council actually concluded after reviewing nearly a century of research. The test works better than chance. It does not work the way its name says it does. And in a question as consequential as whether someone is telling the truth, the distance between “better than chance” and “certain” is not a detail. It is everything.

*Hands on a laptop beside a stethoscope. Guides claiming to teach polygraph countermeasures circulate freely online — one more reason the test's results depend heavily on who is sitting in the chair.* *A phone showing a search screen. Most people meet the polygraph through television and viral claims — not through the research reviews that qualify its reputation.*