Humans perceive expected stimuli faster and more accurately. However, the mechanism behind the integration of expectations with sensory information during perception remains unclear. We investigated the hypothesis that such integration depends on “fusion”—the weighted averaging of different cues informative about stimulus identity. We first trained participants to map a range of tones onto faces spanning a male–female continuum via associative learning. These two features served as expectation and sensory cues to sex, respectively. We then tested specific predictions about the consequences of fusion by manipulating the congruence of these cues in psychophysical and fMRI experiments. Behavioral judgments and patterns of neural activity in auditory association regions revealed fusion of sensory and expectation cues, providing evidence for a precise computational account of how expectations influence perception.