In a study of infant smiling, 20 mother-infant dyads were videotaped in normal face-to-face interaction when the infants were 9 and 14 weeks of age. Videotapes were used to determine which of two classes of smiling behavior models, either arousal modulation or activation indicator, was most supported by empirical data. Arousal modulation models view the infant's smile as a component of a process of modulation reducing cognitive arousal to moderate levels. Activation indicator models, such as the Interaction Task Model, view the smile as an indication that the infant's arousal, activation, or energy levels are rising, either as a result of stimulation or in anticipation of increased cognitive or physical activity. At each data-gathering point, 1 hour of usable interaction was obtained from each dyad, in which the infant was alert and content. Onsets and offsets of smiling, looking, and vocalizing in both infant and mother were coded for time-lagged conditional probability analysis. A set of 11 opposing predictions was established to contrast the models. Opposite conditional probability relationships were predicted in 6 cases, and opposite age differences in conditional relationships were predicted in 5 cases. Reported findings were consistent with the activation indicator models. It is concluded that the Interaction Task Model furthers understanding of early infant social behavior. Numerous line graphs are provided. (RH)