Audiovisual speech illusions are a spectacular illustration of the effect of visual cues on the perception of speech. Because they allow dissociating perception from the physical characteristics of the sensory inputs, these illusions are useful to investigate the cerebral processing of audiovisual speech. However, the meaningless, monosyllabic utterances typically used to induce illusions are far removed from natural communication through speech. We developed naturalistic speech stimuli that embed mismatched auditory and visual cues within grammatically correct sentences to induce illusory perceptions in controlled fashion. Using intracranial EEG, we confirmed that the cortical processing of audiovisual speech recruits an ensemble of areas, from auditory and visual cortices to multisensory and associative regions. Importantly, we were able to resolve which cortical areas are driven more by the auditory or the visual contents of the speech stimulus or by the eventual perceptual report. Our results suggest that higher order sensory and associative areas, rather than early sensory cortices, are key loci for illusory perception. Naturalistic audiovisual speech illusions represent a powerful tool to dissect the specific roles of individual cortical areas in the processing of audiovisual speech.