Many firms incentivize their employees on the basis of subjective performance evaluation by a supervisor. Altruistic or spiteful supervisors may bias performance evaluations, however, which undermines employees’ incentives to exert effort. The present paper studies this agency problem in a dynamic principal-supervisor-agent model. The supervisor is either altruistic or spiteful towards the agent, which is unobservable to the principal. The analysis yields three novel results. First, the supervisor’s compensation is optimally tied to verifiable information about the agent’s performance. Second, in equilibrium, the altruistic supervisor is sometimes intrinsically motivated to give correct performance evaluations, rather than exhibit favoritism. Third, both for sufficiently low and sufficiently high values of the discount factor, the principal attracts only altruistic supervisors.