The use of principal components analysis (PCA) for the study of evoked-response data may be complicated by variations from one trial to another in the latency of underlying brain events. Such variation can come from either random intra-and intersubject variability or from the effects of independent variables that are manipulated between conditions. The effect of such variability is investigated by simulation of these latency-varying events and by analysis of evoked responses in a behavioral task, the Sternberg memory search task, which is well known to generate variation in the latency of brain events. The results of PCA of within-subjects differences in these two situations are plausibly related to underlying stages of information processing, and the technique may augment reaction time data by providing information on the time of occurrence as well as the duration of stages of information processing.