The recent ready availability of recursive partitioning analysis has provided medical investigators with an alternative to the usual parametric multivariate techniques. We believe that recursive partitioning analysis has several distinct advantages: it provides a simple and intuitive method for classifying subjects, it is designed to identify synergistic interactions among factors, it may identify nonlinear relationships with the endpoint, it provides a simple format for constructing homogeneous risk strata, and it considers prior probabilities and penalties for misclassification in its variable selection process. Conversely, parametric multivariate techniques have several advantages over recursive partitioning analysis: they may be better at finding predictive factors during the later stages of the selection process, they may be less subject to the problems of multiple testing and over-training, and they may be better able to account for the full predictive power of a continuous factor. Although neither type of technique is better in all situations, we believe that recursive partitioning analysis will often be the preferred multivariate method and is especially useful for identifying interaction terms that may then be included in parametric multivariate analyses.