We investigate whether and how social ties affect performance in teams by implementing a field experiment in which a sample of undergraduate students are randomly assigned to either teams composed by friends or teams composed by individuals not linked by friendship relationships. Students undertake an intermediate exam divided into two parts: one graded on the basis of individual performance and the other graded on the basis of the team performance. We find that students assigned to socially connected teams perform significantly better than control students in both the team part and the individual part of the exam, suggesting that social ties are relevant both for solving free-riding problems and for inducing knowledge spillovers among teammates. The positive effect of friendship persists over time: treated students obtain better grades also in a second individual test after the conclusion of the experiment.