This paper studies how firms can efficiently incentivize supervisors to truthfully report employee performance. To this end, I develop a dynamic principal-supervisor-agent model. The supervisor is either selfish or altruistic towards the agent, which is observable to the agent but not to the principal. The analysis yields two key results. First, supervisor altruism sometimes provides a net incentive to report performance truthfully, rather than to bias evaluations upward. The intuition is that an altruistic supervisor values his job because of his good relationship with the agent, and puts his job at risk by overrating the agent's performance. Second, I show that by screening for one supervisor type, firms can incentivize the supervisor to truthfully report performance at the lowest possible costs. For this reason, screening may be optimal, even though it reduces the probability that vacancies are filled.