This paper reports the results of experiments designed to test the theory of the optimal composition of prizes in contests. In the aggregate the behavior of subjects is consistent with that predicted by the theory, but we find that such aggregate results mask an unexpected compositional effect on the individual level. Whereas theory predicts that subject efforts are continuous and increasing functions of ability, the actual efforts of our laboratory subjects bifurcate. Lowability workers drop out and exert little or no effort, and high-ability workers try too hard. This bifurcation, which is masked by aggregation, can be explained by assuming loss aversion on the part of the subjects. (JEL: C92, D44, D72, D82, J31) for helpful comments, as well as Maher Said for valuable research assistance. Thanks are also due to seminar participants at Bonn, Copenhagen, Erfurt, Humboldt University Berlin, Rotterdam, and Tilburg.