Polygraph Alternatives:
7 Technologies Compared

Seven technologies are in practical or research use for credibility assessment: polygraph, ocular-motor testing (EyeDetect), voice stress analysis, fMRI, P300 brain fingerprinting, thermal imaging, and FACS-based facial analysis. None of them detects lies. Each measures an indirect physiological correlate of stress, arousal or cognitive load and infers deception from it. The US National Academy of Sciences found polygraph false negative rates of 20–47% in security screening contexts, against roughly 10.2% in specific-incident testing — the gap that drives interest in alternatives. Of the seven, only polygraph and ocular-motor testing are deployed at scale; fMRI and P300 remain research instruments; voice stress analysis performs at chance level in independent trials. They differ on four procurement-relevant axes: how many channels they measure, whether they require a structured test protocol, whether data leaves the site for processing, and whether the output is auditable. The table below scores each technology on all four.

The seven technologies at a glance

The column that matters most in procurement is the last one. Six of the seven technologies were designed for specific-incident testing — a known event, a defined question set, a single subject. Security screening is a different problem: very low base rates of actual deception mean that even a modest false positive rate produces far more incorrect flags than genuine ones. This is the specific failure the National Academy of Sciences identified in polygraph, and it applies to every technology in the table.

1. Polygraph (control question technique)

Polygraph remains the most widely deployed credibility assessment technology. It measures respiratory rate, skin conductance and cardiovascular response as proxies for anxiety during questioning. The US National Academy of Sciences concluded in 2003 that polygraph lacks scientific validity for security screening, with false negative rates reaching 47% in controlled studies.

Its core problem is specificity: elevated physiological arousal is produced by many states other than deception — anxiety about being tested, embarrassment, surprise, anger. Countermeasures such as physical tension or mental arithmetic during control questions can suppress the differential examiners rely on. See our detailed breakdown of the polygraph false negative rate.

2. Ocular-motor testing (Converus EyeDetect)

EyeDetect measures pupillary response, fixation and reading behaviour during a structured computer-based test. It monitors one physiological channel and requires the subject to complete a specific protocol, which makes it impractical for continuous or natural-interaction monitoring. Results are processed on Converus servers in the US, which disqualifies it from SCIF deployment. Vendor-reported accuracy is around 86–90%; independent peer-reviewed replication is limited. See EchoDepth vs EyeDetect for a head-to-head.

3. fMRI-based deception detection

Functional MRI measures blood-oxygen-level-dependent response in prefrontal and parietal regions associated with response inhibition. Laboratory accuracy of roughly 90% has been reported, which is the highest of any technology here — but it has never been validated in field conditions, requires the subject to lie immobile in a multi-million-pound scanner, and was excluded as evidence in US v. Semrau (2010). It is a research instrument, not a deployable capability.

4. Voice stress analysis (VSA / CVSA)

Voice stress analysis claims to detect deception through micro-tremors in speech. Multiple independent assessments — including a 2002 US Department of Defense Polygraph Institute study and subsequent National Institute of Justice field studies — found it performs at chance level. It has no peer-reviewed validation and is not accepted in any procurement context applying evidential standards. It is included here because it is still actively marketed, not because it works.

5. P300 / brain fingerprinting

P300-based methods measure an event-related EEG potential elicited when a subject recognises a probe stimulus — testing for recognition of crime-relevant detail rather than deception as such. Reported accuracy is high but rests on small samples and contested methodology, with minimal independent replication. It requires an EEG cap, a carefully constructed stimulus set, and a specialist operator, and it cannot be applied where the relevant details are already publicly known.

6. Thermal imaging

Thermal imaging measures periorbital blood flow as an arousal proxy. Pavlidis and colleagues reported around 75–80% laboratory accuracy in Nature in 2002. Its advantage is throughput and non-contact operation, making it attractive for border and checkpoint use; its weakness is specificity and sensitivity to ambient temperature, clothing, exertion and recent physical activity, all of which are uncontrolled in exactly the high-throughput settings where it is most attractive.

7. FACS-based facial Action Unit analysis

FACS-based analysis — as implemented in EchoDepth — analyses 44 facial Action Units per frame, including micro-expressions, suppression patterns and AU temporal sequences that are physiologically constrained and difficult to control consciously. Relative to the other six:

• 44 channels simultaneously, against polygraph's 3–4 peripheral signals or EyeDetect's single ocular channel
• Camera-only and non-contact — no sensor attachment, no structured test protocol, works during natural interaction
• Structured timestamped output rather than a subjective examiner report
• On-premise and air-gap capable — see SCIF-compatible AI deployment
• Built on a 40-year peer-reviewed measurement framework, rather than a contested or absent evidence base

Where FACS-based analysis is not the right tool

A comparison that concludes the author wins on every axis is not a comparison. Four situations where FACS-based analysis is the wrong choice:

• You need a binary verdict. EchoDepth does not output truthful/deceptive, and no technology in this table reliably can. If your process requires a yes/no answer, none of these seven will give you a defensible one.
• The face is not visible. Masking, head covering, subject turned away, poor or strongly directional lighting, or low-resolution footage all make FACS coding unusable. Polygraph and VSA are unaffected by this; thermal imaging degrades more gracefully.
• You need standalone legal evidence. UK courts do not admit polygraph, and FACS-based output is not admissible as standalone proof of deception either. It is a structured analytical record that forms part of an evidential picture, nothing more.
• You want unsupervised automated screening. The low-base-rate problem described above applies to FACS-based analysis exactly as it applies to polygraph. Without an analyst interpreting output in context, the false positive burden makes it operationally useless.

The honest summary: FACS-based analysis does not escape the epistemic limit that constrains every technology here — inference from an indirect physiological correlate. Its advantages are channel count, auditability, and the fact that it can be deployed on existing cameras inside an air-gapped environment. Those are real and they are procurement-relevant, but they are not a claim to truth.

How to choose between them

For UK defence, intelligence and security procurement, five criteria separate the seven technologies in practice:

1. Scientific validity — is there an independent, peer-reviewed evidence base for the measurement, separate from vendor accuracy claims?
2. Legal defensibility — does it produce a structured, timestamped, reproducible record that survives review?
3. Operational practicality — does it work during natural interaction, or does it require a structured test, a scanner, or an examiner?
4. Sovereignty — where is the data processed? Cloud dependency rules out SCIF and most HMG deployments.
5. Integration — does output reach SIEM, case management and DSAT-compatible audit records, or does it stop at a PDF?

Scored against these, fMRI and P300 fail on practicality, VSA fails on validity, polygraph fails on validity for screening and on integration, EyeDetect fails on sovereignty and practicality, and thermal imaging fails on specificity. FACS-based analysis meets all five — with the four limitations set out above, which are limits of the problem rather than of the implementation.