In this paper the authors discuss how random forests, ensembles of weakly-correlated decision trees, can be used in concert with fuzzy logic to both classify storm types based on a number of radar-derived storm characteristics and provide a measure of "confidence" in the resulting classifications. The random forest technique provides measures of variable importance and interactions as well as methods for addressing missing data, suggesting fruitful ways to transform the input data and to structure the final classification algorithm. Cross-validation is used as the basis for tuning the algorithm parameters.