A real effort experiment is investigated in which supervisors have to rate the performance of individual workers who in turn receive a bonus payment based on these ratings. We compare a baseline treatment in which supervisors were not restricted in their rating behavior to a forced distribution system in which they had to assign differentiated grades. We find that productivity was significantly higher under a forced distribution by about 8%. But also in the absence of forced distribution, deliberate differentiation positively affected output in subsequent work periods.