Event-related potentials were recorded from normal subjects in an auditory selective attention task. Targets were rare longer (170-ms) tones of a designated pitch, embedded in a sequence of 100-ms standard tones. The effects of attention-modulated processing were evident in the event-related potentials elicited by the standards. Those to relevant standards were similar for easy (1000 Hz vs. 2000 Hz) and hard (1000 Hz vs. 1030 Hz) pitch separations, and were more negative frontocentrally than those to irrelevant standards. Difference waveforms (attended minus unattended standards) revealed Nd, a negative deflection that was earlier in latency for the easy task (onset, 120 ms; peak, 250 ms) than for the hard task (onset, 250 ms; peak, 350 ms). The speed of detection of the deviant longer tones was insensitive to the attention-modulated processes indexed by Nd. Median reaction time did not differ between tasks, although there were more misses and false alarms in the hard task (and nearly all of the latter were to the irrelevant longer tones). Neither direction of attention nor task difficulty affected the latency of mismatch negativity, N2, or P3 (as identified in difference waveforms: attended or unattended longer tones minus their respective standards). The data suggest that performance was guided by two independent but converging processes, automatic mismatch detection of the longer tone and attention-modulated processing of pitch, followed by selection of response.