We argue that a factor widely seen as facilitating cooperation in an international dispute, mediation, is also a sign that a resulting agreement is likely to be short-lived. Mediators get the tough cases, disputes that are most likely to result in short-lived settlements—a selection effect. At the same time, mediation helps to resolve a conflict's underlying issues, making mediated settlements more likely to last—a process effect. We develop a theory that captures these opposing forces—focusing on the conditional effect of mediation on nonstate actors—and taking into account critical aspects of the disputants, conflict, conflict management processes and settlement. Using hazard analysis, Heckman two-stage probit and logistic regression, we statistically analyze over 1,400 settlements drawn from the just released International Conflict Management 2000 data set and also conduct a quantitative case study of mediation tools used in the former Yugoslavia. All analyses find strong support for the opposing effects of mediation; mediated agreements are more likely to be short-lived, unless they involve nonstate actors. This study advances our understanding of selection effects and the factors that lead to short-lived settlements, and provides policy makers and potential mediators a way to establish realistic expectations about the durability of dispute settlements.