We study the role of peer groups in determining the structure and the total amount of executive compensation. Our analysis is based on a standard agency model in which the agent's reservation utility is related to the peer group used for performance evaluation. Our main result is that the informativeness criterion proposed by Holmström (1979) is neither a necessary nor a sufficient condition for the optimality of a relative performance evaluation. Whenever the relative performance evaluation is positively related to the agent's reservation utility, the principal faces a trade-off between the benefits from improved risk sharing and the total cost of compensation. If the peer group effect is strong, it can be optimal to evaluate the agent on her own firm performance only. If the relative performance evaluation is negatively related to the agent's reservation utility, it can also prove useful to reward the agent on the basis of uninformative signals. We also study the optimal weighting and composition of the performance index and find that the principal puts lower (higher) weight on an index and on peer firms that are positively (negatively) related with agent's reservation utility. In case of a negative relation it can even be optimal to include firms with uncorrelated cash flows into the index in order to reduce the total compensation.