I examine optimal incentives and performance measurement in a model where an agent has specific knowledge (in the sense of Jensen and Meckling) about the consequences of his actions for the principal. Contracts can be based both on “input” measures related to the agent's actions and an “output” measure related to the principal's payoff. Whereas input-based pay minimizes income risk, only output-based pay encourages the agent to use his knowledge efficiently. In general, it is optimal to use both kinds of performance measures. The results help to explain some empirical puzzles and lead to several new predictions.