Judges and jurors often rely on consistency for assessing veracity. The present study examined the diagnostic value of within-pair consistency to predict truth-telling in pairs of children aged 8 to 10 years. Twenty-three pairs were questioned about one experienced event and one imagined event (which they had discussed before questioning). Within-pair consistency was significantly higher for experienced events than for imagined events. The diagnostic value of within-pair consistency to predict truth-telling was, however, modest: approximately one out of three judgments based on this cue would have been mistaken. Analyses of children's discussions of the imagined events revealed that interview questions about topics that had been discussed before questioning did not effectively discriminate experienced and imagined events, providing support for theoretical assumptions underlying the unanticipated-question approach. Practical recommendations for police interviewers are provided.