“Forecast targeting,” forward-looking monetary policy that uses central-bank judgment to construct optimal policy projections of the target variables and the instrument rate, may perform substantially better than monetary policy that disregards judgment and follows a given instrument rule. This is demonstrated in a few examples for two empirical models of the U.S. economy, one forward looking and one backward looking. A practical finite-horizon approximation is used. Optimal policy projections corresponding to the optimal policy under commitment in a timeless perspective can easily be constructed. The whole projection path of the instrument rate is more important than the current instrument setting. The resulting reduced-form reaction function for the current instrument rate is a very complex function of all inputs in the monetary-policy decision process, including the central bank’s judgment. It cannot be summarized as a simple reaction function such as a Taylor rule. Fortunately, it need not be made explicit.