This study examines the hazards of democracy promotion in multi-ethnic societies. I begin by developing a simple formal model, which isolates the main features of the strategic context of democracy promotion that can trigger ethnic conflict. The key intuition of the model is that democracy promotion creates a commitment problem between ethnic majorities and minority groups that are demographically over-represented in the coercive forces of the authoritarian regime that has been removed to install democracy. Specifically, the majority cannot credibly commit to these groups not to purge them from the coercive forces of the future regime creating strong incentives for them to mobilize while they still have control over the means of coercion of the state. Based on this intuition, the model predicts that minority groups that are demographically over-represented in the coercive forces of the authoritarian regime--their share of the manpower of these forces exceeds their population share--are more likely to rebel against the new regime imposed by the democracy-promoter than minority groups that are demographically under-represented. Further, the more demographically over-represented the demographically over-represented minority group the more likely it is to rebel. I denote this proposition the warrior's curse. I test my warrior's curse prediction in the empirical context of transitions from British colonial rule post-World War I, when democracies emerged, not through endogenous processes, but by imperial diktat. Using an original dataset on the pre-colonial, colonial, and post-colonial status of 168 minority groups in 29 ex-British colonies, I find two sets of novel empirical results. First, I show that minority groups who were demographically over-represented in colonial police forces, or the so-called 'martial races, ' were no more likely to have the 'habits' of fighting in the pre-colonial period than demographically under-represented minority groups. On the contrary, demographic over-representation in colonial police forces is negatively correlated with the presence of centralized political institutions in the pre-colonial period and positively correlated with the presence of some form of slavery in the pre-colonial period. Second, controlling for these proximate determinants of demographic over-representation in colonial police forces, and several other confounding factors, I find robust support for the warrior's curse. The propensity for armed rebellions during decolonization, a period defined as the first 12 years after independence, turns out to be strongest among the demographically over-represented minorities, and is increasing in the degree of over-representation. Moreover, these results also extend to the intensity of minority rebellions. As a further test of my theoretical model, I also present a focused analytic narrative on the political mobilization by the Sikhs, a martial race of British India, in the province of Punjab as it was being partitioned into Indian and Pakistani Punjab on the eve of the independence of these countries on August 1947. The narrative demonstrates that the Sikh mobilization was motivated by the rational fear that certain key policy decisions of the departing colonial administration had forever undermined the chances of Mohammad Ali Jinnah conceding them weighted representation in the armed forces if they agreed to join Pakistan.