This is a game-theoretic analysis of the link between regime type and international conflict. Democratic leaders can credibly be punished for bad conflict outcomes, whereas autocratic leaders cannot. Due to the fear of being thrown out of office, democratic leaders are (i) more selective about the wars they initiate and (ii) on average win more of the wars they start. Foreign policy behaviour is found to display strategic complementarities. Therefore, the likelihood of interstate war is lowest in the democratic dyad (pair), highest in the autocratic dyad with the mixed dyad in between. The results are consistent with empirical findings.