Recently there has been a great deal of interest in studying monetary policy under model uncertainty. We develop new methods to analyze different sources of uncertainty in one coherent structure, which is useful for policy decisions. We show how to estimate the size of the uncertainty based on time series data, and how to incorporate this uncertainty in choosing policy. In particular, we develop a new approach for modeling uncertainty called model error modeling. The approach imposes additional structure on the errors of an estimated model, and builds a statistical description of the uncertainty around the model. We develop both parametric and nonparametric specifications of this approach, and use them to estimate uncertainty in a small model of the US economy. We then use our estimates to compute Bayesian and minimax robust policy rules, which are designed to perform well in the face of uncertainty.