Five experiments (N = 2,204) examined responses to a realistic moral dilemma: a military pilot must decide whether to bomb a dangerous enemy target, also killing a bystander. Few people endorsed bombing when the bystander was an innocent civilian; however, when the bystander's identity was unknown, over twice as many people endorsed the bombing. Follow-up studies tested boundary conditions and found the effect to extend beyond modern-day conflicts in the Middle East, showing a similar pattern of judgment for a fictional war. Bombing endorsement was predicted by attitudes towards total war, the theory that there should be no distinction between military and civilian targets in wartime conflict. Bombing endorsement was lower for UK compared to US participants due to differences in total war attitudes. This work has implications for conflicts where unidentified bystanders are common by revealing a potentially deadly bias: people often assume unidentified bystanders are guilty unless proven innocent.