Visual speech cues presented in synchrony with discrepant auditory speech cues are usually combined to a surprisingly clear unitary percept that corresponds with neither of the two sensory inputs (the McGurk illusion). This audiovisual integration process is commonly believed to be highly autonomous and robust to cognitive intervention, unlike the processing of ambiguous phonemes which has been shown to be dependent on lexical-semantic context and other higher cognitive variables. To investigate this issue, three experiments were carried out in which subjects' expectations were varied as they were presented stimuli containing the McGurk effect. In Experiments 1 and 2, the illusion was embedded in real words that were presented in semantically congruent vs. incongruent sentential contexts. In Experiment 3, nonlexical stimuli containing the McGurk illusion either matched or did not match subjects' prior expectations. Results show that the clarity of the illusion, and to some extent the probability of the illusion, was significantly influenced by subjects' expectations. Thus perceptions that are based on audiovisually integrated speech cues are not immune to cognitive influences; rather, they seem to be subject to the same functions and variations as ambiguous phonemes.