The mismatch negativity (MMN) is an event-related brain potential elicited by infrequent, physically deviant sounds in a sequence of repetitive auditory stimuli. Two dichotic listening experiments that were designed to optimize the selective focusing of attention provided a strong test of Näätänen's proposal that the MMN is unaffected by attention and reflects the operation of a strongly automatic mismatch detection system. In Experiment 1, tones were presented at intervals of 120-320 ms, and the deviant tones (intensity decrements) in both the attended and unattended ears elicited negative waves consistent in waveshape, latency, and distribution with previously described MMNs. In contrast to previous reports, however, the MMN elicited by the unattended-channel deviant was markedly reduced (peak amplitude of less than 1 microV) relative to the corresponding negative wave elicited by the attended-channel deviants (3-4 microV), as well as relative to previously reported MMNs (3-6 microV) elicited by comparable deviations in stimulus intensity. In Experiment 2, which employed interstimulus intervals of 65-205 ms, the unattended-channel MMN elicited by the deviant fainter tones was barely discernible, whereas the corresponding attended-channel negativity was again about 3-4 microV. These findings call into question the assertion that the auditory mismatch detection process and the associated MMN wave are wholly independent of attentional influence. Rather, these data provide evidence that the processing of stimuli in unattended channels can be attenuated or gated at an early sensory level under conditions of highly focused auditory selective attention.