We analyze a multiple-activity, principal-agent model in which the activities are naturally substitutable for the agent and complementary for the principal. A basic result is that the optimal compensation must cause the agent to view the activities as complements. This complementarity is achieved by employing a compensation scheme that is typically nonmonotone and makes success on multiple dimensions the sole source of large rewards. A number of empirical implications follow, along with explanations for some existing empirical findings. We also discuss applications to compensation in specific occupations.