Kenwick, Vasquez, and Powers contend that defensive alliances provoke conflict; Leeds and Johnson argue that they deter conflict. I use a formal model to deduce conditions under which alliances provoke and those under which they deter, allowing us to distinguish the two effects. The model shows that both sets of analyses include cases that should be left out of an analysis of whether alliances are more likely to provoke or deter. I re-analyze both sides' data and find evidence suggestive that alliances provoke only when the two sides do not have a recent history of conflict or alliances but deter when they do.