Prior studies of mother-infant interaction have generally used a variable-centered approach to associate face-to-face communication with psychosocial outcomes. Herein, we use a person-centered approach to identify clusters of infants who exhibit similar behavioral profiles during face-to-face communication with their mothers. Four infant communication channels were examined--gaze, facial affect, vocal affect, and head orientation--coded from videotape at a 1-s temporal resolution. We used k-means clustering to classify community infants (N = 132) into 10 groups, based on variation in the intercept and the autocorrelation function at the first time lag, representing respectively the overall level of behavior and the predictability of the infant's moment-by-moment behavioral stream, in each of the 4 communication channels. In this exploratory study, 10 clusters were identified, some with unusual levels or predictability of behavior in varying channels, and clusters associated differentially with risk outcomes (infant 4-month temperament and 12-month attachment). Distinct forms of affective dysregulation were identified: sustained negative vocal affect associated with degree of disorganization; random vocal affect associated with attachment resistance; random facial affect and vocal affect, irrespective of positive/negative valence, associated with infant difficult temperament. Clustering multiple channels of infant communication generated unique behavioral profiles and predicted 4- and 12-month outcomes, suggesting that these clusters may indeed represent natural types of infant communicative behavior, not easily observed with the naked eye, that may be useful behavioral markers of clinical risk.