Arms races are costly and inefficient; therefore, standard causal explanations, based on threat perception, are inadequate, as states should prefer to resolve disagreements prior to the onset of these inefficient competitions. Building on recent research, arms races are alternatively conceptualized as a product of uncertainty, used to reveal information. Expectations are derived regarding when arms races should be most likely, allowing for one of the first systematic, quantitative tests of the causes of arms races. Empirical tests support theoretical expectations that arms races are most likely in contexts where there are salient competitive stakes and high levels of uncertainty, such as territorial rivalries or the early tenure of new leaders. The theoretical logic and empirical tests not only produce insights into why states participate in costly and inefficient arms races but also hint at a better understanding of the long puzzling relationship between arms races and war.