We addressed the question of whether the mismatch negativity (MMN) event-related potential reflects an attention-independent process. Previous studies have shown that the MMN response to intensity deviation was significantly reduced or even abolished when attention was highly focused on a concurrent sound channel, whereas no conclusive evidence of attentional sensitivity has been obtained for frequency deviation. We tested a new hypothesis suggesting that competition between detection of identical deviations in attended and unattended channels and the biasing of this competition induced by the subject's task account for the observed MMN effects. In a fast-paced dichotic paradigm, we set up competition for frequency MMN and removed it for intensity MMN. We found that frequency MMN was now abolished in the unattended channel, whereas the amplitude of the intensity MMN was unaffected. These results support the competition hypothesis and suggest that selective attention in and of itself does not affect the MMN. Top-down processes can determine what information reaches the deviance-detection process when changes in multiple channels vie for the same MMN resource and one of the competing changes is relevant for the subject's task.