In our simple model the supervisor: i) cannot observe the agent's effort; ii) aims at inducing the agent to exert high effort; but iii) can only offer rewards based on performance. Since performance is only stochastically related to effort, evaluation errors may occur. In particular, deserving agents that have exerted high effort may not be rewarded (Type I errors) and undeserving agents that have exerted low effort may be rewarded (Type II errors). We show that, although the model predicts both errors to be equally detrimental to performance, this prediction fails with a lab experiment. In fact, failing to reward deserving agents is significantly more detrimental than rewarding undeserving agents. We discuss our result in the light of some economic and managerial theories of behavior. Our result may have interesting implications for strategic human resource management and personnel economics and may also contribute to the debate about incentives and organizational performance.