Field evidence suggests that agents belonging to the same group tend to behave similarly, i.e., behavior exhibits "social interaction effects". Testing for such effects is difficult because of severe identification problems. To isolate social interaction effects in a way that avoids these problems, we design an experiment where each subject simultaneously is a member of two randomly assigned groups with different group members. In both groups subjects play exactly the same public goods game. In our data we isolate "social interaction effects", i.e., a majority of subjects is very strongly influenced by the contributions of the respective group members.