We empirically investigate possible distortions in subjective performance evaluations. A key hypothesis is that evaluations are more upward biased the closer the social ties between supervisor and appraised employee. We test this hypothesis with a company data set from a call center organization which contains not only subjective assessments but also several more objective measures of performance. Controlling for these performance measures, we find strong evidence that evaluations are upwards biased in smaller teams and some evidence that supervisors give better ratings to employees they themselves have evaluated before.