The ability of 3-month-old infants to learn arbitrary auditory–visual associations between voices and faces was investigated by familiarizing each infant to two alternating stimuli presented on a VCR monitor. Each stimulus was a voice–face combination, where the voices and faces were male and/or female. On the post-familiarization test trials each infant was presented alternately with a familiar and a novel voice–face combination, where the novel combination consisted of a voice and a face they had heard and seen previously (but not together), and on these test trials attention was significantly higher to the novel combination. These findings are a clear demonstration that 3-month-olds can learn arbitrary voice–face associations, and they are discussed in terms of early intermodal perception and face perception. Copyright © 2001 John Wiley & Sons, Ltd.