Research on the accuracy of eyewitness testimony has expanded dramatically in recent years. Most of it concerns the issue of mistaken identification, not the problem of uncovering deceptive accounts of witnesses, which is the focus of this research. In the literature, a technique for lie detection has been proposed that induces cognitive load on liars by averting their rehearsal of deception: Time Restricted Integrity-Confirmation. The current authors tested it by instructing "witnesses" of actual crime videos to lie or tell the truth to related questions. Each of 145 adults was randomly assigned to a truth telling, an unrehearsed lying, or a rehearsed lying condition. The cognitive cues were response time, answer consistency, eye movements, and pupil dilation. Eye data were gathered with an infrared eye tracker. Truth tellers had the quickest response times and the fewest inconsistencies. Moreover, they generally had more eye movements, suggesting low cognitive loads. Discriminant analyses classified rehearsed liars, unrehearsed liars, and truth tellers up to 69% accurately, with few false positives. Further refinement is warranted.