In this research, the development of intermodal perception in infancy was examined by using a new method, the intermodal learning method. 3-month-old infants were given the opportunity to learn a relation between 2 single film and soundtrack pairs through a 2-min familiarization period under 1 of 4 conditions. Films of naturalistic events were accompanied by a soundtrack that was (1) appropriate to the composition of the object and synchronous with its motions, (2) appropriate and nonsynchronous, (3) inappropriate and synchronous, or (4) inappropriate and nonsynchronous. A group of control subjects was familiarized with irrelevant films and soundtracks. Then all subjects were tested in a 2-choice intermodal preference test to determine under which familiarization conditions intermodal learning had occurred. Results indicated that only subjects who had been familiarized with appropriate and synchronous film and soundtrack pairs showed evidence of intermodal learning as compared with the performance of control subjects. Intermodal learning occurred on the basis of 2 kinds of invariant audio-visual relations, temporal synchrony, and temporal microstructure specifying the composition of the object. Intermodal learning did not occur through association on the basis of co-occurrence, nor did it occur when any incongruent audio-visual structure was present. These findings support an invariant-detection view of the development of intermodal perception.