Eyewitness identification research has reliably shown that accurate identifications are faster than inaccurate identifications. Recently, D. Dunning and S. Perretta (2002) claimed that an identification latency of 10-12 s not only best discriminates between accurate and inaccurate identifications but also produces extremely high accuracy rates, approaching 90%. Consistent with predictions from recognition memory theory, however, we show that the optimum time boundary varies with overall response latency under manipulations of retention interval and nominal lineup size, and that the accuracy rate inside the optimum time boundary is much less impressive than previously reported. We outline directions for clarifying the accuracy and latency relationship to assist the reliable diagnosis of identification accuracy.