In a selective attention experiment, event-related potentials (ERPs) were recorded from 12 subjects who listened to sequences of high and low frequency tone pips, presented binaurally in random order at a rapid rate. Their task was to attend selectively to either the high or low frequency tones and to press a button upon detection of occasional longer-duration tones at the appropriate frequency. The effect of attention was manifest in difference wave forms obtained by subtracting ERPs to the shorter tones when they were unattended from their ERPs to the same stimuli when they were attended. The difference waves consisted of a broad, multiphasic negative component (Nd) with its later phase (300-400 msec) being more frontally distributed on the scalp than either its early phase (100-300 msec) or the N1 component that was evoked in a reading condition. The amplitude and latency of onset of Nd varied systematically as a function of the frequency separation between the tones. It was concluded that the effect of selective auditory attention on the N1 component (ca. 100 msec latency) is not due solely to an enlargement of the exogenous N1 component of the vertex potential, but rather includes the addition of a prolonged endogenous component. A simple gating mechanism for human selective attention was thus rejected.