Since Taylor's 1993 paper researchers have devoted a lot effort to estimation of monetary policy rules. Taylor showed that a simple central bank reaction function, with the interest rate as monetary policy instrument and inflation and output gap as explanatory variables, mimics the Fed funds rate pretty well during the period from 1987 to 1992. Often, the Taylor rule coefficients are interpreted as if they reflect central bank's preferences. However, this may be misleading. In this paper we show that Taylor rule coefficients are complicated terms consisting of preference parameters as well as parameters given by the structure of the economy. We illustrate our conclusion that Taylor rule coefficients cannot be interpreted as reflecting central bank preferences by estimating standard forward-looking Taylor rules for the Bundesbank, the Fed and UK and confront these with our results obtained by a multi-equation GMM approach in order to detect central bank preferences.