Methods of optimizing the Lovejoy and Austin technique to delineate areas of precipitation using visible and infrared satellite data are investigated. The technique involves training the satellite data by correlation with real-time radar data. The choice of statistical measures to define the precipitation/no-precipitation boundary between satellite classes is investigated. Subjective evaluation of the satellite-diagnosed precipitation fields indicates that minimizing the difference between the observed and diagnosed number of precipitation pixels produces the most realistic results. Maximization of some standard skill scores tends to overestimate the areas extent of the precipitation. Examples of the variability of the accuracy of the technique and the variation in optimum boundary or threshold are given. Cases illustrating the improvement produced by using different correlation tables for different synoptic systems are presented. Use of time-averaged correlation tables is investigated and found to be nearly as accurate as use of tables formed at one time, when evaluated within the training area. Fixed predefined tables were rather less accurate when evaluated within the training area, especially with respect to the diagnosed areal extent. A method is presented to combine the use of instantaneous and time-averaged correlation tables, together with a predefined table. Ideally, the method should incorporate the use of different correlation tables for different synoptic systems.