If civil war is a contest for popular support, why would a government ever embark on a policy of disproportionate force and mass killing? The logic of civilian defection expects such an approach to easily backfire, as civilians respond to massive losses by opposing the side that inflicted them. Yet even if civilians balance against the side they believe most likely to kill them, massive violence can still occur and – when it does – is sometimes seen as a key source of success. Using an epidemic model of popular support dynamics, this paper derives a set of conditions under which mass killing can occur in civil war. Such conditions emerge when combatants have imperfect information about civilian defection, when learning is slow, or when one side initially enjoys an asymmetric advantage in levels of active support or flows of recruits.