Abstract: In large-N investigations, civil conflicts – like any significant political event – tend to be studied and understood at the country level. Popular explanations of why and where civil wars occur, however, refer to such factors as ethnic discrimination, wealth inequalities, access to contrabands, and peripheral havens. The intensity of such factors varies geographically within states. Therefore, any statistical study of civil war that uses country-level approximations of local phenomena is potentially flawed. In this paper, we disaggregate the country and let 100x100 km grid cells be the units of observation. Having developed geo-referenced conflict data from Uppsala/PRIO’s conflict database, we use GIS to identify regions of peace and conflict and as a tool to generate sub-national measures of key explanatory variables. The results from an empirical analysis of African civil wars, 1970–2001, demonstrate spatial clustering of conflict that co-varies with the spatial distribution of several exogenous factors. Territorial conflict is more likely in sparsely populated regions near the state border, at a distance from the capital, and without significant rough terrain. Conflict over state governance is more likely in regions that are densely populated, near diamond fields, and near the capital city. These promising findings show the value of the innovative research design and offer nuanced explanations of the correlates of civil war.