Swedish nonsense syllables distinguished solely by their vowels [i], [y] or [e], were presented to phonetically sophisticated subjects auditorily, visually and in cross-dubbed audiovisual form with incongruent cues to openness, roundedness or both. Acoustic [y] dubbed onto optic [i] or [e] was heard as a retracted [i], while acoustic [i] or [e] dubbed onto optic [y] were perceived as rounded and slightly fronted. This confirms the higher weight of the more reliable information and that intermodal integration occurs at the level of phonetically informative properties prior to any categorization.