Despite the large amount of empirical research on monetary policy rules, there is surprisingly little consensus on the nature or even the existence of changes in the conduct of U.S. monetary policy. Three issues appear central to this disagreement: (1) the specific type of changes in the policy coefficients, (2) the treatment of heteroskedasticity, and (3) the real-time nature of the data used. This paper addresses these issues in the context of forwardlooking Taylor rules with drifting coefficients. The estimation is based on real-time data and accounts for the presence of heteroskedasticity in the policy shock. The findings suggest important but gradual changes in the rule coefficients, not adequately captured by the usual split-sample estimation. In contrast to Orphanides (2002, 2003), I find that the Fed's response to the real-time forecast of inflation was weak in the second half of the 1970s, perhaps not satisfying Taylor's principle as suggested by Clarida, Galì, and Gertler (2000). However, the response to inflation was strong before 1973 and gradually regained strength from the early 1980s onward. Moreover, as in Orphanides (2003), the Fed's response to real activity fell substantially and lastingly during the 1970s.