This paper models two key roles of subjective performance evaluations: their incentive role and their feedback role. The paper shows that the feedback role makes subjective pay feasible even without repeated interaction, as long as there exists some verifiable measure of performance. It also shows that while subjective pay is helpful, it cannot achieve full efficiency. However, fully efficient incentives are achievable if the firm can commit to a forced distribution of evaluations and employs a continuum of workers. With a small number of workers, a forced distribution is valuable only if the verifiable measure is poor.