In this study, we examined the predictive power of confidence in within-subject terms. Ninety-six subjects answered, and made confidence ratings for, a series of questions about a crime they witnessed. The average between-subjects and within-subject accuracy-confidence correlations were comparably low. Confidence is neither a useful predictor of the accuracy of a particular witness nor of the accuracy of particular statements made by the same witness. Another possible predictor of accuracy, response latency, correlated only negligibly with accuracy, but more strongly with confidence. This pattern was obtained for both between-subjects and within-subject comparisons. The theoretical and practical emplications of these results are discussed. (Author abstract)