Reported correlations between accuracy and certainty of eyewitness identifications are sometimes positive but equally often nil. The predictability failure of memory accuracy by means of witness confidence can be explained by looking at factors that affect memory but not necessarily confidence (i.e., conditions of witnessing, inconsistent information during the retention interval, labeling, stereotypes) and factors that affect confidence but not necessarily memory (i.e., non-forced-choice recognition tests, biased testing instruction, interrogation, and implicit beliefs that facial memory is good). Research into these factors shows that integrative memorial and cognitive processes occur unconsciously and therefore may alter either memory or confidence independently of each other. As a result, accuracy-confidence correspondence should be inversely related to the extensiveness of reconstructive memory processes (which change memory but not confidence) or suggestive social influences (which change confidence but not memory). Noncorrespondence is expected when memory is altered by inconsistent information, a criminal stereotype, or a descriptive label of the suspect or when confidence is altered by factors that promote commitment to testimony or trust in facial memory. Police and lawyers should avoid behaviors that facilitate these effects and, along with jurors and possibly even witnesses, should be informed that confidence is often a poor index of accuracy. A note and 27 references are cited. (Author abstract modified)