In this paper we analyze disinflation policy when a central bank has imperfect information about private sector inflation expectations but learns about them from economic outcomes, which are in part the result of the disinflation policy itself. The form of uncertainty is manifested as uncertainty about the effect of past disinflation policy on the current output gap. This differs from other studies on learning and control in a monetary policy context (e.g., [Ellison, 2006] and [Svensson and Williams, 2007]) that assume uncertainty about the effects of current policy actions on the economy. We derive the central bank's optimal disinflation strategy under active learning (DOP) and compare it with two limiting cases—certainty equivalence policy (CEP), or passive learning, and a Brainard-style cautionary monetary policy (CP). It turns out that under the DOP inflation stays between the levels implied by the CEP and the CP. A novel result—e.g., unlike Beck and Wieland (2002)—is that this holds irrespective of the initial level of inflation. At high levels of inherited inflation the DOP moves closer to the CEP, at low levels of inherited inflation the DOP resembles the CP.