Currently, there are two opposing views on feature binding in the auditory modality: according to behavioral studies, this process requires focused attention, whereas electrophysiological studies suggest that feature binding may be fully automatic and independent of attention. Here, we examined whether feature binding depends on higher-level attentional processes by manipulating the attentional focus. We used four auditory stimuli that differed in two features: pitch and location. Two rare deviants could be detected within a sequence of two frequent standards exclusively by feature conjunctions rather than by any single feature alone. Event-related potentials to auditory stimuli were analyzed for four conditions: selective attention to target auditory deviants, selective ignoring of nontarget auditory deviants, nonselective distributed attention to all stimuli within auditory modality, and selective attention diverted from auditory to visual modality. The negative difference (Nd) between event-related potentials to deviants and standards was measured within two time intervals, corresponding to mismatch negativity (100-200 ms) and N2b (200-300 ms). Only under the condition of selective attention to specific feature conjunctions, prominent Nd was observed in mismatch negativity as well in N2b time ranges, whereas no significant Nd was observed in other conditions. As Nd is considered a marker of deviance processing, our results support the view that deviance was not detected unless attention was focused on the stimuli, thus supporting the view that feature binding requires attention.