Should people be allowed to leave joint projects freely or should they be deterred from breaking off? This depends on why people stop collaborating and whether they have good reasons to do so. We explore the factors that lead to the breakdown of partnerships by studying a public good game with imperfect public monitoring and an exit option. In our experiment, subjects were assigned a partner with whom they could contribute over several periods to a public good with stochastic out- comes. They could choose in each period between participating in the public project or working on their own. We find there was excessive exit especially because subjects over-estimated the likelihood their partner would leave. Treatments with high barriers to exit generated higher welfare overall as they fostered stability and prevented inefficient breakdowns in relationships. There were differences across treatments in the intensity with which different factors drove the choice to work alone. Differences in expected payoffs between independent and group work were more important as a driver of exit in treatments with low barriers to exit. The intensity of other factors was more constant across treatments, including whether the common project failed in the previous period, the belief that one’s partner did not want to maintain the partnership and the belief that he exerted less effort than oneself.