Biometric Accuracy Metrics Explained: FAR, FRR, and EER

Every biometric system makes a probabilistic decision, not a perfect one, and False Acceptance Rate (FAR), False Rejection Rate (FRR), and Equal Error Rate (EER) are the standard metrics used to describe how often it gets that decision wrong. Understanding these metrics is essential for comparing biometric technologies and setting the right sensitivity for a given application.

False Acceptance Rate measures how often the system incorrectly matches two different people's biometric samples, granting access to someone who should have been rejected. False Rejection Rate measures the opposite error: how often the system fails to match a legitimate person's own biometric sample against their own enrolled template, denying access to someone who should have been accepted. Neither error rate can be driven to zero simultaneously with the other using a single fixed threshold, which is the core tension every biometric deployment must navigate.

Biometric matching produces a similarity score, and the system compares that score against a decision threshold to decide match or no-match. Lowering the threshold makes the system more lenient, reducing FRR (fewer legitimate users rejected) but increasing FAR (more impostors accepted). Raising the threshold does the opposite. The right threshold depends entirely on the application's risk tolerance.

Because FAR and FRR move in opposite directions as the threshold changes, comparing two biometric systems using only one of these numbers can be misleading, since a vendor could report an impressively low FAR simply by setting a very strict threshold that also produces an unusably high FRR. Equal Error Rate solves this by identifying the single threshold at which FAR and FRR are equal, providing one number that summarizes a system's overall discriminative power independent of a specific deployment's threshold choice. A lower EER generally indicates a more accurate underlying biometric algorithm.

• High-security access (nuclear facilities, government vaults): prioritize very low FAR, accepting higher FRR and occasional legitimate user friction
• Consumer convenience (phone unlock): prioritize low FRR so the device feels responsive, accepting a somewhat higher FAR since a PIN fallback exists
• Large-scale identification (border control, criminal databases): FAR matters enormously because the system compares against millions of records, where even a small FAR produces many false candidate matches

Real-world evaluations also track Failure to Enroll Rate (people whose biometric cannot be captured well enough to create a usable template) and Failure to Acquire Rate (a live capture attempt that produces an unusable sample), both of which affect real-world usability independent of the matching algorithm's raw accuracy. A system with excellent EER but a high Failure to Enroll Rate can still be a poor practical choice for a population with, for example, many manual laborers whose fingerprints are frequently worn down.