Injurious aggression in group housed male laboratory mice is a common welfare issue that can also negatively affect study outcomes. Often, one mouse in the cage appears unwounded, and the current standard practice is to remove this presumed aggressor. This procedure is not based on empirical evidence and may impede welfare by singly housing animals. We experimentally tested the hypothesis that the apparently uninjured mouse is indeed the aggressor, and that aggression is reduced in his absence. We separated cages of four or five male mice, reported for fight wounds to our university’s veterinary service, into cages of two or three mice containing either only wounded mice (“wounded” treatment) or both wounded and unwounded mice (“mixed” treatment). We recorded aggressive behavior for 30 min immediately pre- and post-separation, and scored wound severity at separation and over two weeks after. We predicted that if unwounded mice are aggressors: mice in the wounded treatment would show less escalated aggression (involving biting) than mice in the mixed treatment, and would be wounded less and/or heal faster during the two weeks following separation. Wound scores decreased significantly after separation in both treatments (wounded: p < 0.0001; mixed: p = 0.011), but mice in the wounded treatment healed faster than those in the mixed treatment (p = 0.006). There was no significant effect of treatment on duration of escalated aggression in the 30 min following separation (p = 0.240), nor did treatment predict which cages would be re-separated due to continued aggression (p = 0.104). Our results support the hypothesis that the unwounded mouse is the aggressor, as mice in cages with an unwounded mouse healed more slowly than those without. Both types of groups healed significantly over time, suggesting that separation into groups of two or three is a possible management alternative to social isolation of the presumed aggressor. By identifying spontaneous cases of severe aggression in an existing colony, we obtained a heterogeneous and representative sample of clinical cases, bolstering the generalizability of our conclusions.