Three major groups of models of sexual selection differ in their predictions about the relation between degree of ornamentation and future survival prospects of males, provided that the phenotypic expression of ornament size is condition-dependent and that condition has a genetic component. The `good genes' models predict a positive relation between ornamentation and survival, while neither the `arbitrary traits' nor `good parent' models predict this relation. I evaluated this prediction in a natural population of swallows Hirundo rustica, which are sexually size dimorphic in tail length. Previous experiments have shown that females prefer males with extravagant tail ornaments. Male survivors in this study had significantly longer tails than non-survivors, irrespective of age and mating status. By contrast, this was not the case for females. Survival selection in males on all other morphological characters examined was weak and statistically non-significant, showing that selection on tail length was not the result of selection on correlated characters. The selection differential for tail length due to survival selection was high, particularly among unmated and yearling male swallows. My results thus provide support for a `good genes' model of sexual selection rather than the alternative models.