Can we predict when and where violence will break out within cases of genocide? Given often weak political will to respond, knowing where to strategically prioritize limited resources is valuable information for international decision makers contemplating intervention. I develop a theoretical model to help identify areas vulnerable to violence during genocide. I argue vulnerability is a function of the state's coercive power and the ruling elite's control of this power from above, mediated by the strength of society's cohesion below. Violence will be delayed in areas where political and military resistance to the center is high as it takes time for extremists to exert control at the periphery. Violence will also be delayed in well-integrated communities as it takes time to break existing social bonds and destroy social capital. I draw on the case of Rwanda's 1994 genocide and examine sub-national variation in the onset of violence across the country's 145 administrative communes using survival analysis and within-case analyses comparing early and late onset in two communes. The findings have implications for international policy makers responding to ongoing genocides.