An overview of the principle feature subset selection methods is given. We investigate a number of measures of feature subset quality, using large commercial databases. We develop an entropic measure, based upon the information gain approach used within ID3 and C4.5 to build trees, which is shown to give the best performance over our databases. This measure is used within a simple feature subset selection algorithm and the technique is used to generate subsets of high quality features from the databases. A simulated annealing based data mining technique is presented and applied to the databases. The performance using all features is compared to that achieved using the subset selected by our algorithm. We show that a substantial reduction in the number of features may be achieved together with an improvement in the performance of our data mining system. We also present a modification of the data mining algorithm, which allows it to simultaneously search for promising feature subsets and high quality rules. The effect of varying the generality level of the desired pattern is also investigated.