A third of police facial search users wrong or unsure whether a match counts as evidence, survey finds

Police officers and staff in England and Wales who use the national police database's face search mostly rate it highly, but many are unclear about what a match can prove, and most have not been trained on it recently, according to a Home Office survey published on November 12, 2026. The department's researchers surveyed users of the facial search function of the Police National Database (PND) between April 8 and June 2, 2026, and received 779 valid responses.

The tool is a form of retrospective facial recognition. It compares an image of an unknown person, typically taken from CCTV or phone footage, with custody photos held by police; a Home Office summary from December 2025 put the number searchable at about 19 million. Every force in England and Wales has access to it. About 6,000 police users ran around 25,000 searches a month in 2025, the report says. The software returns a list of candidates that pass a similarity threshold, a trained operator then compares the faces, and a result may be used only as an investigative lead, not as evidence.

What users get wrong

Most respondents had the basics right. In the survey, 93% knew that a match is not definitive confirmation of identity, and 86% knew they should consider more than the top-ranked candidate. But around 34% either wrongly believed that a match meets the evidential requirement for recognition under PACE Code D, the code that governs how police identify people, or were unsure. Frequent users, those running 60 or more searches a year, were more likely to answer correctly than infrequent users, by 71% to 62%.

While 86% recalled some training or guidance on the tool, around 65% could not recall any in the past year, and 14% did not recall ever receiving it. Among those who recalled training, 89% had been shown how to submit searches, but only 45% how to prepare images and 49% what the tool's risks and limitations are. Just over a third (36%) said they would benefit from more.

The sample has limits. Taking part was voluntary, and the 779 valid responses amount to 13% of the 6,012 people who used the tool between February 2025 and January 2026. The response rate was 36% among those invited as frequent users and 10% among those invited as infrequent ones.

Accuracy and bias

The algorithm used in the database was tested by the National Physical Laboratory (NPL), and the Home Office published the results in December 2025. At the setting police use, the system returned a match in over 99% of searches where the person sought was in the database, the Home Office summary says, but in a limited set of circumstances it was more likely to wrongly include some demographic groups in its results.

In the survey, 65% of users said they think the tool performs differently for different groups to some extent. Among those who thought so, 74% said a subject's ethnicity affects its performance at least somewhat, more than said the same of age (59%) or gender (48%). The report stresses that these answers show how users see the tool and are not an objective assessment of its accuracy or bias.

What happens next

In a November 4 announcement, the Home Office said it has independently tested a new algorithm for retrospective searches that proved “highly accurate and without bias,” and that forces will be able to roll it out from next year. Its December 2025 summary of the NPL test had already said that a new algorithm had been procured and independently tested, could be used at settings with no significant demographic variation in performance, and was due for operational testing early in 2026. The same December publication included an NPL evaluation of an Idemia algorithm intended for a new national facial matching service; the Home Office pages we read do not say whether that is the algorithm forces will roll out. The November 4 announcement also said that technologies used by law enforcement will have to be independently tested for bias under a planned law, as we reported.

The survey was commissioned to inform an inspection of how police forces and the National Crime Agency use retrospective facial recognition, which the Home Secretary ordered from His Majesty's Inspectorate of Constabulary and Fire & Rescue Services on November 12, 2025. Its findings were shared with the inspectors. The Home Office publication does not include the inspection's own conclusions.