This paper applies regime-switching methods to the problem of measuring monetary policy. Policy preferences and structural factors are specified parametrically as independent Markov processes. Interaction between the structural and preference parameters in the policy rule serves to identify the two processes. The estimates uncover policy episodes that are initiated by switches to "dove regimes," shown to Granger-cause both NBER recessions and the Romer dates. These episodes imply real effects of monetary policy that are smaller than those found in previous studies.