Play has proved to have a central role in children's development, most notably in rule learning (Piaget, 1965; Sutton-Smith, 1979) and negotiation of roles and goals (Garvey, 1974; Bruner et al., 1976). Yet very little research has been done on early play. The present study focuses on early social games, i.e., vocal-kinetic play routines that mothers use to interact with infants from very early on. We explored 3-month-old infants and their mothers performing a routine game first in the usual way, then in two violated conditions: without gestures and without sound. The aim of the study is to investigate infants' participation and expectations in the game and whether this participation is affected by changes in the multimodal format of the game. Infants' facial expressions, gaze, and body movements were coded to measure levels of engagement and affective state across the three conditions. Results showed a significant decrease in Limbs Movements and expressions of Positive Affect, an increase in Gaze Away and in Stunned Expression when the game structure was violated. These results indicate that the violated game conditions were experienced as less engaging, either because of an unexpected break in the established joint routine, or simply because they were weaker versions of the same game. Overall, our results suggest that structured, multimodal play routines may constitute interactional contexts that only work as integrated units of auditory and motor resources, representing early communicative contexts which prepare the ground for later, more complex multimodal interactions, such as verbal exchanges.