A fundamental aspect of our sensory experience is that information from different modalities is often seamlessly integrated into a unified percept. Recent computational and behavioral studies have shown that humans combine sensory cues according to a statically optimal scheme derived from Bayesian probability theory; they perform better when two sensory cues are combined. We have explored multisensory cue integration for self-motion (heading) perception using both visual (optic flow) and vestibular (linear acceleration) signals. We recorded from single neurons in the dorsal medial superior temporal area (MSTd) of visual cortex during a heading discrimination task where trained monkeys, like humans, behaviorally combine visual and vestibular cues to improve heading perception. Under bimodal stimulation, MSTd neurons combine visual and vestibular cues linearly but sub-additively. Neurons with congruent heading preferences for visual and vestibular stimuli show improved sensitivity and lower neuronal thresholds under cue combination. In contrast, neurons with opposite preferences show diminished sensitivity under cue combination. We further show that MSTd responses are significantly correlated with the monkeys' perceptual decisions in a congruency-dependent manner. Deficits in behavior brought by chemical inactivation of this area provide further support of the hypothesis that extrastriate visual cortex mediates multisensory integration for self-motion perception. These findings provide the first behavioral demonstration of statistically-optimal cue integration in non-human primates and identify a population of neurons that may form its neural basis.