This paper proposes a new framework to analyse systematic and unsystematic monetary policy within the same econometric model. We model explicitely two important features of the policy decision process: current economic activity is measured with error and policy makers have access to information from a large number of data series. This framework is then exploited to evaluate US monetary polich from 1983 to 2001. We find that (i) two aggregate shocks explain the bulk of economic activity; (ii) systematic policy differs depending on whether it faces demand or technology shocks; (iii) monetary policy shock has virtually no effect on real variables and the expected sign on prices and interest rate.