Numbers of civilian deaths and injuries have become important measurements of conflict violence. This article analyzes the methodological and epistemological underpinnings of reports on civilian casualties by the US Forces and the United Nations Assistance Mission in Afghanistan (UNAMA) as technologies of quantification. We argue that the quantification of conflict violence relies on – but obscures – judgments about the boundaries of the phenomenon that is being counted and the evidence that is required to confirm incidents. In particular, the competing civilian casualty counts operate on the basis different concepts of who counts as a civilian, what counts as conflict violence, and what counts of evidence of civilian casualties. We illustrate this argument with four examples: the distinction between direct and indirect deaths, the boundary between civilians and non-civilians, the boundary between conflict violence and criminal violence, and hierarchies in the visibility of civilians. We also draw attention to the racialized hierarchies of credibility that shape the US military’s dismissal of the voices of Afghan survivors and witnesses of conflict violence. Civilian casualty counts are products of specific methods, epistemologies, standards of proofs, and definitions. They help constitute an understanding of armed conflict as quantifiable and manageable.