Why do some militias perpetrate terrible abuses against civilians and others rarely do? While existing research suggests that governments delegate extreme or brutal violence to militias, this argument cannot account for the patterns of violence by states and militias, nor for the remarkable variation in perpetration of violence across militias. In this analysis, we analyze one type of civilian abuse that should be especially likely to be delegated by states to militias: sexual violence. Using an original dataset, we examine the patterns and determinants of conflict-related sexual violence by all African militias active during civil conflicts, and in the immediate post-conflict period, between 1989-2009. We find that in every case of militia-perpetrated sexual violence, the state also was reported to be a perpetrator, presenting a strong challenge to previous arguments about the delegation of violence. We then analyze predictors of militia-perpetrated sexual violence by examining three levels of potential determinants: state-level, conflict-level and group-level. We find that low state capacity and the recruitment of children are both strong predictors of militia-perpetrated sexual violence. We interpret this finding as support for an argument that armed groups that operate in states with low capacity, and those with low social cohesion, are more likely to perpetrate sexual violence. In addition, contrary to common assumptions, we find lethal violence and sexual violence do not covary. Through exploring these patterns, this paper can begin to shed light on questions of relevance to both scholars and policymakers, and especially the question of whether states appear to be “outsourcing” violence. More broadly, the study of militia-perpetrated sexual violence can offer important insight into those forms of violence that may be considered gratuitous or excessively brutal.