Understanding the conditions under which state leaders are willing to honor alliance commitments in war will increase knowledge about the escalation and diffusion of conflict and about the propensity of states to fulfill agreements under anarchy. New data analysis provides evidence that alliance commitments are fulfilled about 75 percent of the time. But how can one understand the failure of alliance partners to act as promised in the remaining 25 percent of cases? Formal modelers have deduced that because of the costs associated with alliances, state leaders who form alliances are likely to fulfill them; those alliances that are formed should be fairly reliable. I argue, therefore, that one can best account for violations of alliance agreements either through an understanding of the factors that reduce the costs of violation or through changes that have occurred since the alliance was formed. Using detailed data on alliance commitments between 1816 and 1944, I find evidence commensurate with this argument. Changes in the power of states or in their policymaking processes are powerful predictors of the failure to honor past commitments; and nondemocratic states and major powers, sets of states that I argue suffer lower costs from reneging on agreements, are more likely to violate treaties.