In contexts such as education and sports, skill-accumulation of individuals over time crucially depends on the amount of training they receive, which is often allocated on the basis of repeated selection. We analyze optimal selection policies in a model of endogenous skill formation where, apart from their ability to transform training into skills, individuals also differ with respect to relative age. The latter has been identified by recent empirical research as a major determinant for performance differentials within cohorts. We find that the optimal policy is pro-competitive at later selection stages in the sense of selecting the individuals with the higher skill signals. All eventual corrections due to relative age occur at early stages, where selection is either counter-competitive (i.e. individuals with low skill signals are selected) or even avoided at all. Thereby, the induced selection quality is non-monotone in the degree of ex-ante asymmetry due to relative age. Finally, the (empirical) observation of persistent relative age effects does in general not hint at suboptimal selection policies.