Adult eye direction was manipulated while adults interacted with 3-6 month-olds over closed-circuit television (Experiment 1) or in person (Experiment 2). Infants received 4 1-min interaction periods. For experimental groups, adult eye contact was maintained during Periods 1 and 3, and averted during Periods 2 and 4 (by viewing infants on a television monitor to maintain contingency). Control infants received eye contact during all periods. Experimental infants' smiling declined whenever adults looked away; their visual attention simply decreased across periods. Control infants showed little change in gaze or smiling across periods. The implications of these results for Baron-Cohen's model of infant theory of mind and Morton and Johnson's 2-process theory of infant face perception are discussed.