This paper considers a general class of nonlinear rational-expectations models in which policymakers seek to maximize an objective function that may be household expected utility. We show how to derive a target criterion that is: (i) consistent with the model's structural equations, (ii) strong enough to imply a unique equilibrium, and (iii) optimal, in the sense that a commitment to adjust the policy instrument at all dates so as to satisfy the target criterion maximizes the objective function. The proposed optimal target criterion is a linear equation that must be satisfied by the projected paths of certain economically relevant "target variables." It takes the same form at all times and generally involves only a small number of target variables, regardless of the size and complexity of the model. While the projected path of the economy requires information about the current state, the target criterion itself can be stated without reference to a complete description of the state of the world. We illustrate the application of the method to a nonlinear DSGE model with staggered price-setting, in which the objective of policy is to maximize household expected utility.