A polygraph doesn’t detect lies. That’s worth stating plainly before anything else, because it’s the single most misunderstood fact about a machine most people have strong opinions about. What a polygraph actually measures is stress — and the entire, century-long controversy around “lie detector” tests comes down to one uncomfortable fact: lying can cause stress, but so can almost everything else that happens in an interrogation room.

What a polygraph actually measures

The physiological logic is straightforward. A polygraph records several bodily signals simultaneously while a subject answers a structured series of questions: blood pressure and pulse, breathing patterns (measured with pneumographs around the chest), and galvanic skin response — essentially how much you’re sweating, tracked through the skin’s electrical conductivity. The underlying theory is that deception triggers measurable stress, and stress produces these physiological changes, so a spike across these signals during a specific question is treated as a marker of possible deception.

The problem is built directly into that logic. Sweating, an elevated heart rate and changed breathing aren’t unique to lying — anxiety, fear, anger, embarrassment, and the simple stress of being wrongly accused of something you didn’t do can all produce the exact same physiological signature. The machine can tell you a subject is stressed. It cannot, on its own, tell you why.

What the actual accuracy research shows

A landmark 2003 report from the National Academy of Sciences remains the most authoritative independent review of polygraph validity, and its conclusions were considerably less favorable than the confident “90% accurate” figures that circulate informally. The report found that the standard Comparison Question Test format used in most polygraph exams could identify deception correctly roughly 70% of the time under controlled research conditions — a real effect, better than random chance, but far short of the reliability implied by treating a polygraph result as definitive evidence. The report also concluded that the broader scientific basis for CQT polygraph testing was weak, and that a meaningful share of the existing research supporting it was low quality.

The more consequential problem isn’t the overall accuracy number — it’s the direction the errors run. The American Psychological Association has specifically warned that the rate of false positives — truthful people incorrectly flagged as deceptive — is “unacceptable.” Because the test can’t distinguish the stress of lying from the stress of simply being interrogated, an anxious, entirely honest person can fail a polygraph, while a genuinely deceptive person who is calm, practiced, or trained in countermeasures can pass one. That asymmetry is the core reason courts have treated the technology with such consistent skepticism for a century.

Why courts don’t trust it — a legal history that starts with the polygraph itself

The polygraph’s relationship with the American legal system is unusually direct: the foundational US legal standard for admitting any scientific evidence originates from a polygraph case. Frye v. United States, decided in 1923, involved an early precursor to the modern polygraph and established what became known as the Frye Standard — expert scientific evidence is admissible only if the underlying technique is generally accepted as valid within its relevant scientific community. The polygraph failed that test in 1923, and general scientific opinion about its reliability hasn’t shifted enough since then to change the outcome.

The Frye Standard governed federal courts until 1993, when the Supreme Court replaced it with a more detailed reliability framework in Daubert v. Merrell Dow Pharmaceuticals, asking judges to weigh whether a technique has been tested, peer-reviewed, has a known and acceptable error rate, and is generally accepted in its field. Polygraph evidence has struggled under both standards for the same underlying reason: neither is satisfied by a technique a meaningful share of the scientific community still doubts. In 1998, the Supreme Court issued its most direct ruling on the subject in United States v. Scheffer, upholding a rule that categorically excluded polygraph evidence from military courts-martial — a decision that reflects, rather than creates, the technology’s shaky scientific standing.

The one meaningful exception is the stipulated polygraph: cases where both the prosecution and defense agree in advance that a specific test’s results will be admissible, regardless of the outcome. That mutual, upfront agreement — not any change in the underlying science — remains the most common way polygraph evidence actually makes it into a modern American courtroom.

So why do polygraphs still get used constantly?

Despite failing the courtroom admissibility bar for a century, polygraphs remain widely used in pre-employment screening (particularly for law enforcement and national security roles), internal investigations, and parole or probation monitoring — contexts where the legal rules of evidence simply don’t apply. Part of the technology’s real-world value isn’t the physiological measurement at all — it’s the psychological effect of believing you’re being tested. A subject who thinks the machine works is more likely to feel genuine stress when answering deceptively, and more likely to confess or contradict themselves under the pressure of the exam itself, independent of whether the machine’s readings are scientifically definitive. That’s a real, documented use case — but it’s a very different claim from “this machine detects lies,” and conflating the two is exactly where public understanding of the technology tends to go wrong.

The actual takeaway

A polygraph is a real, functioning piece of physiological measurement equipment, not a fraud in the sense of measuring nothing at all — but it measures stress, not deception, and the gap between those two things is wide enough that courts have excluded the technology as a matter of settled legal doctrine since the same case that defined how American courts evaluate scientific evidence in the first place. Roughly 70% accuracy under ideal research conditions, an “unacceptable” false-positive rate by the American Psychological Association’s own assessment, and a century of consistent judicial skepticism are the actual scientific and legal record — considerably more modest than the cultural image of a machine that reliably catches liars.

For the other major (and similarly overstated) lie-detection method, see are micro-expressions real, and can you actually be trained to read them?.