We present an empirical investigation of video-mediated free play between 13 pairs of friends (ages 7 and 8). The pairs spent 10 minutes playing with each of four different prototypes we developed to support free play over videoconferencing. We coded each interaction for the types of play and the amount of social play observed. The children in our study were largely successful in playing together across videoconferencing, though challenges in managing visibility, attention, and intersubjectivity made it more difficult than face-to-face play. We also found that our prototypes supported some types of play to varying degrees. Our contribution lies in identifying these design tradeoffs and providing directions for future design of video-mediated communication systems for children.