In recent years, subgroup analysis has emerged as an important task due to inherent heterogeneity of the subjects in clinical trials and market segmentation analysis. However, most of the existing work for subgroup analysis has considered a small number of candidate variables for characterizing the subgroup membership. In the data-rich era, the candidate variables could be high dimensional such as genetic data of patients in clinical trials. In this talk, we consider a high dimensional mixture model to jointly capture the subgroup membership as well as the within-subgroup behavior and develop a Bayesian variable selection method in that framework. We investigate the theoretical and empirical properties of the proposed method, and suggest how model-based inference may be carried out in subgroup analysis.