Model uncertainty has the potential to change importantly how monetary policy should be conducted, making it an issue that central banks cannot ignore. In this paper, I use a standard new Keynesian business cycle model to analyze the behavior of a central bank that conducts policy with discretion while fearing that its model is misspecified. I begin by showing how to solve linear-quadratic robust Markov-perfect Stackelberg problems where the leader fears that private agents form expectations using the misspecified model. Next, I exploit the connection between robust control and uncertainty aversion to present and interpret my results in terms of the distorted beliefs held by the central bank, households, and firms. My main results are as follows. First, the central bank's pessimism leads it to forecast future outcomes using an expectations operator that, relative to rational expectations, assigns greater probability to extreme inflation and consumption outcomes. Second, the central bank's skepticism about its model causes it to move forcefully to stabilize inflation following shocks. Finally, even in the absence of misspecification, policy loss can be improved if the central bank implements a robust policy.