The purpose of this study was to determine whether certain previously reported relationships between event-related potentials and measures of signal detection performance occur in vigilance as well as psychophysical settings. In the course of the study, evidence was found which challenges previously proposed psychological correlates of "P300". EEG was recorded while 15 subjects carried out a 40-min signal detection test. CNV was measured between a warning click and the brief offset of a dim light denoting the possible (p = 0.5) occurrence of the signal, a faint tone in the constant background noise. P300 was measured at a point 300 msec following offset of the dim light. As reported previously in psychophysical settings, P300 amplitude was positively related to signal intensity and response confidence, and was larger for correct detections (Hits) than for correct rejections, misses or false alarms. From first to second half of the test both Hits and false alarms fell, response criterion beta rose, and the amplitude of both CNV and P300 fell. The latter negative relationship between beta and P300 contrasted with a positive one when subjects rated their signal reports at three levels of confidence; here the most confident ratings (high beta) were associated with the highest amplitude of P300. CNV reflected individual ability to sustain performance. Results are explained in terms of a two-factor version (Wilkinson, 1976) of the prior state/reactive change hypothesis (Karlin, 1970). It is suggested that time on task constitutes a prior state influence on P300, whereas other variables influence P300 by means of either reactive change or 'real' change in an endogenous P300 component.