The EEG was monitored during an auditory vigilance task. The wanted signal consisted of three consecutive odd digits. Five other types of signal varied in their approximation to the wanted signal and were ranked on that criterion for their ‘arousal’ value. Each signal was followed by a rest period. The key results were: (i) alpha abundance (8·5–13·5 Hz) diminished as the arousal value of the signal increased (P < 0·005); (ii) theta (4·5–6·5 Hz) and beta (13·5–19·5 Hz) showed either weak or no effects; (iii) a very-low-frequency filter (2·0–4·5 Hz) yielded a strong effect for one class of signal only (P < 0·021); (iv) during rest periods which followed the three most arousing types of signal, alpha abundance was greater than during the preceding signal (P < 0·025): the reverse held for the least arousing signals (P < 0·025); (v) theta abundance was very much lower following response (rest period following the wanted signal) than following other signal types (P < 0·01) and theta abundance during rest was generally smaller than during signals (P < 0·021); (vi) for the three least arousing signals, activity at 2·0–4·5 Hz was lower during rest than during signals (P < 0·05); (vii) subjects rated themselves as being more ‘keyed up’ as cue or arousal value of the signals increased (P < 0·001) and (viii) similarly, reported that alertness increased during rest periods associated with signals of increasing cue value (P < 0·01). These findings demonstrate that during a task of this sort, the subject experiences variation in alertness; this variation in subjective state is reflected by systematic change in the EEG. It is concluded that these results lend support to arousal models of vigilance and to Lindsley's activation theory.