The computer system contained facial features that allowed the generation of more than 34 billion composites of faces. The witness initially rated 20 possible faces for their general resemblance to a suspect. These ratings were then used to guide a genetic algorithm search process which generated a new set of 20 faces based on the most highly rated composites of the first set. The witness rated the new set, and the process continued until the offender's face was shown. A binary coding system was found to be the most efficient, and simulations showed that a close resemblance to the offender's face could be developed in 10 rounds of ratings. The process was evaluated experimentally using 120 student volunteers and three hypothetical offenders. The students developed the facial composites immediately, 3 days, or 7 days after viewing a videotape of a simulated armed robbery. Results showed that face recall varied with the offender's facial characteristics. Prototypical faces were poorly recognized, but faces with distinctive features were well remembered up to 7 days after exposure. The computer process (FacePrints) evolved composites that were identified by independent raters more than half the time. Figures, photographs, and 43 references