This paper confirms the existence of peer effects in a learning process with data from an experiment. The experimental setting offers an insight into the mechanisms of peer interaction and provides complementary information to empirical studies using survey or administrative data. The results show that a partner has a motivational effect even before the actual cooperation takes place. The evidence for optimal group composition is not robust. Some of the "better" students improve the performance of their partner but they induce lower motivation. (Contains 1 figure and 6 tables.)