Incentive regimes not only shape workers’ incentives, but also induce worker sorting. Past research finds that workers sorting into high-powered, contingent or competitive regimes tend to be relatively productive. The reasons for higher productivity, however, have not yet been fully understood. These sorted workers tend to have higher raw skills, but also differ on attitudes, preferences and behavioral inclinations. This paper presents results of a field experiment which allow us to precisely distinguish the effect of sorting on the basis of preferences and attitudes — “tastes” for the regime — from the effects of sorting on the basis of raw skills. Further, we contrast these sorting effects with that of varying formal cash incentives. The experimental context is a competitive on-line contest in which elite software developers solve a challenging algorithmic problem over 10 days, competing in independent groups of 20. The effect of sorting on taste, holding raw skills constant, was large. Problem-solving performance doubled when comparing sorted workers with those who were precisely matched on skills but otherwise unsorted. The effect is entirely explained by the higher effort (hours worked) of those sorting on taste, rather than unobserved skills. This effect was roughly the same magnitude as varying the cash prize of a group from $0 to $1000. The effect of skills-based sorting was an order of magnitude smaller than that of sorting on taste. Thus the behavioral implications of sorting were considerably larger than the “compositional” implications of sorting on raw skills and human capital in the classical sense.