We examined the effect of asking unanticipated interview questions on lie detection. We set up a room in which truth tellers (N = 40) performed five tasks. Liars (N = 40) observed the activity room through a window and were asked to generate a story in which they claim to have performed five tasks. The interview schedule included four thematically related pairs of questions. The first question of each pairing was easy for interviewees to anticipate, as it reflected how events are typically recalled. However, the second included a perspective shift (temporal or spatial), which made the question harder to anticipate. We also measured cognitive load via performance on an object sorting secondary task that interviewees undertook during interview. Compared with truth tellers, liars' verbal responses showed a greater decline in details across three out of four question pairs. Liars also sorted significantly fewer objects per minute on the secondary task. Copyright © 2012 John Wiley & Sons, Ltd.