In the wild, house mice live in social groups, whereas in the laboratory male mice are often singly housed. Environmental enrichment such as that provided by social housing has been argued to improve the cognitive performance of laboratory animals in experimental tests. The aim of the present study was to test the cost of aggressive social interactions on learning in male CD-1 mice. We found that subordinate mice from more aggressive dyads showed spatial learning impairment, measured as alternation on a T-maze. Learning impairments in subordinates have hitherto been presumed attributable to the animals' exposure to, and relative standing within, the social group. By contrast, the impairment we observed could not have been the result of recent social defeat because it persisted weeks later when the mice were housed alone. Elevated urinary corticosterone predicted later subordination, though paradoxically these abnormally high levels were reduced by pair housing.