Peer-assisted learning (PAL) can have both benefits and costs when compared to individual learning. In an empirical study aimed at testing whether PAL yields better learning outcomes than individual learning in a set of insight problems from the Remote Associates Test (RAT), participants in the PAL condition learned from interacting with scripted peers and outperformed participants in the individual learning condition. Although interaction was limited, there is evidence for knowledge spillover from peers as a cognitive mechanism in PAL. The peer-assisted learning transferred to a final test that was completed individually. The benefit of peer-assisted learning persisted in spite of significantly greater error exposure in the PAL condition. In addition, the superiority of peer-assisted learning over individual learning was more pronounced for the more difficult RAT problems. These findings suggest that PAL may be an effective approach to developing insight problem solving skills.