I propose a discrete choice method for estimating monetary policy reaction functions based on research by Hu and Phillips (2004). This method distinguishes between determining the underlying desired rate which drives policy rate changes and actually implementing interest rate changes. The method is applied to ECB rate setting between 1999 and 2010 by estimating a forward-looking Taylor rule on a monthly basis using real-time data drawn from the Survey of Professional Forecasters. All parameters are estimated significantly and with the expected sign. Including the period of financial turmoil in the sample delivers a less aggressive policy rule as the ECB was constrained by the lower bound on nominal interest rates. The ECB's non-standard measures helped to circumvent that constraint on monetary policy, however. For the pre-turmoil sample, the discrete choice model's estimated desired policy rate is more aggressive and less gradual than least squares estimates of the same rule specification. This is explained by the fact that the discrete choice model takes account of the fact that central banks change interest rates by discrete amounts. An advantage of using discrete choice models is that probabilities are attached to the different outcomes of every interest rate setting meeting. These probabilities correlate fairly well with the probabilities derived from surveys among commercial bank economists.