When exogenously imposed, rank-order tournaments have incentive properties but their overall efficiency is reduced by a high variance in performance (Bull, Schotter, and Weigelt 1987). However, since the efficiency of performance-related pay is attributable both to its incentive effect and to its selection effect among employees (Lazear, 2000), it is important to investigate the ex ante sorting effect of tournaments. This paper reports results from an experiment analyzing whether allowing subjects to self-select into different payment schemes helps in reducing the variability of performance in tournaments. We show that when the subjects choose to enter a tournament, the average effort is higher and the between-subject variance is substantially lower than when the same payment scheme is imposed. Mainly based on risk aversion, sorting is efficiency-enhancing since it increases the homogeneity of the contestants. We suggest that the flexibility of the labor market is an important condition for a higher efficiency of relative performance pay.