This paper studies a moral hazard problem in which the principal does not know the agent's beliefs about the output generating process. The agent is risk neutral, transfers are subject to limited liability, and the principal evaluates contracts according to their worst‐case payoff against a rich set of plausible agent beliefs. With common knowledge of the relationship between effort and expected output, optimal contracts are of the form w(y)=max(αy + β,0). With or without common knowledge of that relationship, there are broad conditions under which optimal contracts are of the form w(y)=αy + β and the principal cannot improve her payoff guarantee by randomizing over menus of contracts.