Recent experiments show that once a false identification is made, it is virtually undetectable unless discredited by some other form of evidence. Furthermore, false identifications are seldom associated with a striking resemblence between the true offender and the falsely identified person. It is argued that eyewitnesses are prone to choose the lineup member who most resembles the perpetrator relative to other lineup members. This theory is misleading because of the unpredictable occurrence of target-absent lineups and is not corrected fully by instructions to eyewitnesses. A logical extension of this relative judgment theory would be that an inverse relationship exists between the accuracy of witnesses' memories and witnesses' tendencies to rely on relative judgment. To test this hypothesis, 192 students who were told the experiment concerned performance on video games witnessed the theft of a video game as they entered a cubicle. Eyewitnesses were not debriefed, but believed that the theft was real throughout the lineup identification phase. In a control group, 48 eyewitnesses were shown the perpetrator-present lineup and 48 the perpetrator-absent lineup. The remaining 96 witnesses first were presented with the blank lineup and told that the suspect might not be present. They were then asked to look at a second lineup, with the same instructions, and treated the same as the control group. The blank lineup yielded a diagnostic split of eyewitnesses: those who made no identification when presented with a blank lineup were less likely to make false identifications on the actual lineup than either witnesses who identified someone from the blank lineup or those not presented with a blank lineup. Tables and 39 references are supplied. (Author abstract modified)