Mugshot searches fail because the witness must be shown too many photo
s. Current methods to reduce the number rely too heavily on verbal rep
orts of individual features. This research reports three tests of a ne
w system that augments the presently used tools by having the witness
choose photos that are subjectively similar to the target. Each photo
thus chosen increases the ranking of every photo in the album that is
similar to it, as determined by a similarity network in which the albu
m photos are embedded. Because the ranking of the target itself is usu
ally thus increased, it is soon displayed. The tests used familiar tar
gets, an incidental one, and unfamiliar videoed ones. From 76% to 84%
of the ''witnesses'' reached the target, having viewed 2.4 to 8.5 time
s fewer photos than they would have had they simply leafed through pho
tos as is traditionally done.