Current understanding of feature binding remains controversial. Studies involving mismatch negativity (MMN) measurement show a low level of binding, whereas behavioral experiments suggest a higher level. We examined the possibility that the two levels of feature binding coexist and may be shown within one experiment. The electroencephalogram was recorded while participants were engaged in an auditory two-alternative choice task, which was a combination of the oddball and the condensation tasks. Two types of deviant target stimuli were used – complex stimuli, which required feature conjunction to be identified, and simple stimuli, which differed from standard stimuli in a single feature. Two behavioral outcomes – correct responses and errors – were analyzed separately. Responses to complex stimuli were slower and less accurate than responses to simple stimuli. MMN was prominent and its amplitude was similar for both simple and complex stimuli, whereas the respective stimuli differed from standards in a single feature or two features respectively. Errors in response only to complex stimuli were associated with decreased MMN amplitude. P300 amplitude was greater for complex stimuli than for simple stimuli. Our data are compatible with the explanation that feature binding in auditory modality depends on two concurrent levels of processing. We speculate that the earlier level related to MMN generation is an essential and critical stage. Yet, a later analysis is also carried out, affecting P300 amplitude and response time. The current findings provide resolution to conflicting views on the nature of feature binding and show that feature binding is a distributed multilevel process.