We consider how a principal can use randomized strategies in designing optimal contracts in agency settings. We distinguish between ex post randomization (overfee schedulesfollowing act selection by the agent) and ex ante randomization (overfee schedules before act selection). We show that ex ante randomization may be efficient in both full information and private information settings under certain assumptions regarding preferences and the production technology. It is particularly significant that additive separability and concavity ofthe agent's utility function as well as concavity of the production function do not rule out efficient ex ante randomized contracts in private information settings, although, as is known, they rule out ex post randomization.