Social interactions have important consequences for labour market outcomes. Yet the growing literature has relied on indirect definitions of networks. We present the first evidence based on direct information on friends' networks. We address issues of correlated effects with instrumental variables and panel data. We find large network effects, which persist even after controlling for family networks. One additional employed friend increases a person's job finding probability by approximately 13 percent. This is a result of endogenous social interactions. We also provide the first evidence that network effects operate through information transmission rather than through social norms or leisure complementarities.