This project introduces a new framework for thinking about ethnic cleansing in civil war and tests this theory on the Lebanese civil war of 1975-1990 using an original quantitative data set with information on demographic, migration and violence for over 1,400 villages or urban neighborhoods as well as about 60 interviews gathered during 14 months of fieldwork. Ethnic cleansing can be a powerful military strategy for winning a quick and decisive victory yet I argue that, absent political goals such as secession, armed groups have strong political and economic incentives to moderate the amount of wartime displacement. For these reasons armed groups have an incentive to acquire more fine-grained information about resident non-coethnics to allow displacement of only those who are actively disloyal. This is easiest in areas where different ethnic groups mix socially, such as mixed residential neighbourhoods, which are therefore more likely to witness selective violence but less likely to experience full collective ethnic cleansing. This hypothesis is the direct opposite of the conventional wisdom yet finds strong support in the data which stems from voter registration rolls, government documentation, NGO reports, and other sources and allows for a rich set of quantitative tests.