The question of alliance termination raises a variety of theoretical and empirical issues in the study of alliances. While it has been overlooked by many IR scholars, the recent theoretical literature has accumulated a series of new findings by treating alliance itself as the unit of analysis. (Bennett 1997; Leeds and Savun 2007; Morrow 1991) However, this “alliance approach” to the problem leaves crucial questions unanswered: Why do states abandon security agreements that they once welcomed? Under what conditions do states violate them? To answer these empirical questions, it is imperative to focus on the individual state as the unit of analysis for further investigation of state behavior in given international security and domestic contexts. This article, which employs the “state approach,” aims to offer insight into alliance violation. It relies on the Alliance Treaty Obligation and Provisions (ATOP) dataset and uses a competing-risks event-history model for estimating the risks of other termination modes, as in Leeds and Savun (2007). This study reveals the following: First, the statistical findings, which are significant, contradict the predictions of the capability aggregation model. Furthermore, it discusses these findings in light of the suggestion that states opportunistically violate alliances at a higher level of external threats for their own survival, in some cases, or for their aggressive policies, in others. Second, it finds that the effect of domestic politics on a state’s alliance violation is not as salient as external security and national capability factors. Third, the article shows that wartime alliance is short-lived, which is consistent with our intuition, but its termination by violation occurs less frequently than other termination modes. This is because the casus foederis clearly exists for the allies, and, as such, they have less incentive to deviate from a given agreement during wartime.