How do societies break down into violence? While the question of why civil wars start has received considerable attention, the question of how they start has been largely ignored. How does violence spread directly following onset? Building from a novel typology of onset, this article argues that while in some civil wars, violence at onset is geographically concentrated, in other conflicts, it spreads quickly across the territory. This is due to baseline differentials in state capacity: violence spreads faster when the state collapses at onset, typically following incomplete past conflict resolution. Using geocoded data from the Uppsala Conflict Data Program (GED), I compare the spread of violence following onset across civil wars. I find that onset violence is especially likely to be geographically scattered when civil war follows state dissolution, taking the form of widespread communal violence. Only in civil wars that start as a peripheral challenge does violence tend to be concentrated. These findings challenge the widely held assumption that all conflicts necessarily entail a common process of onset and escalation of violence, by providing evidence that civil wars in fact vary in patterned ways. This in turn as far-reaching consequences for research on the causes and dynamics of civil war.