In this paper, we describe the analysis of multimodal data collected on small collaborative learning groups. In a previous study, we asked pairs (N=84) with no programming experience to program a robot to solve a series of mazes. The quality of the dyad's collaboration was evaluated, and two interventions were implemented to support collaborative learning. In the current study, we present the analysis of Kinect[TM] and speech data gathered on dyads during the programming task. We first show how certain movements and patterns of gestures correlate positively with collaboration and learning gains. We next use clustering algorithms to find prototypical body positions of participants and relate amount of time spent in certain postures with learning gains as in Schneider & Blikstein's work. Finally, we examine measures of proxemics and physical orientation within the dyads to explore how to detect good collaboration. We discuss the relevance of these results to designing and assessing collaborative small group activities and outline future work related to other collected sensor data. [For the full proceedings, see ED593090.]