Collaboration is an effective means of learning new information, but is collaboration productive in domains where collaborators hold qualitatively different conceptions of the domain’s causal structure? We explored this question in the domain of evolutionary biology. College undergraduates (n = 44) completed an assessment of their evolutionary reasoning by themselves (pretest), with a partner (dyad test), and several months later (posttest). Collaboration proved more effective for lower-scoring partners than higher-scoring partners. Lower-scoring partners not only increased their score from pretest to dyad test but also maintained higher scores at posttest. Follow-up analyses revealed that, for lower-scoring partners, the smaller the difference between their pretest score and their partner’s pretest score, the greater the gain from pretest to posttest. Bridging the gap between a novice’s view of a conceptually complex domain and an expert’s view appears to require instruction that is initially more aligned with the former than the latter.