Eyewitness memory is widely believed to be unreliable because (a) high-confidence eyewitness misidentifications played a role in over 70% of the now more than 300 DNA exonerations of wrongfully convicted men and women, (b) forensically relevant laboratory studies have often reported a weak relationship between eyewitness confidence and accuracy, and (c) memory is sufficiently malleable that, not infrequently, people (including eyewitnesses) can be led to remember events differently from the way the events actually happened. In light of such evidence, many researchers agree that confidence statements made by eyewitnesses in a court of law (in particular, the high confidence they often express at trial) should be discounted, if not disregarded altogether. But what about confidence statements made by eyewitnesses at the time of the initial identification (e.g., from a lineup), before there is much opportunity for memory contamination to occur? A considerable body of recent empirical work suggests that confidence may be a highly reliable indicator of accuracy at that time, which fits with longstanding theoretical models of recognition memory. Counterintuitively, an appreciation of this fact could do more to protect innocent defendants from being wrongfully convicted than any other eyewitness identification reform that has been proposed to date. (PsycINFO Database Record (c) 2015 APA, all rights reserved).