My initial research effort was in the area of tree-structured regression modeling for censored survival data. With tree-structured regression modeling, the response data are stratified according to particular covariate values, and separate regression models are fit to each stratum. Stratification is performed recursively, using a combination of statistical tests and residual analysis. The significance of this work lies in the fact that it provides a formal goodness-of-fit test for the regression models and keeps the fitted models as simple as possible, for ease of interpretation. I developed Tree-structured proportional hazards regression and tree-structured parametric regression models for censored survival data with Wei-Yin Loh at the University of Wisconsin-Madison. At the National Center for Toxicological Research, I worked on over-dispersed binomial regression tree models with James Chen. In teratology studies, a dose-response model is often fit to bioassay data to provide a relationship between the probability of a developmental defect and the level of exposure. The regression tree approach can be applied to dose response modeling of a developmental effect. Recently I studied extra-Poisson regression tree procedure and applied it to investigate geographic variability in mortality rates on lung cancer in Missouri as well as effects of various demographic variability. The data contain lung-cancer mortality among the 115 counties during the period of 1972-1981 for both sexes. Software is available for the tree-structured regression models.