With rapid advances in understanding of human diseases, the paradigm of medicine shifts from “one-fits-all” to targeted therapies. In targeted therapy development, the patient population is thought of as a mixture of two or more subgroups that may derive differential treatment efficacy. To ?nd the right patient population for the therapy to target, inference on treatment ef?cacy in subgroups as well as in the overall mixture population are all of interest. In this presentation, I will start by introducing two fundamental statistical issues in this inference procedure, followed by establishing a general logical estimation principle. Finally, as a step forward in patient targeting, we present a simultaneous inference procedure based on confidence intervals to logically infer treatment efficacy in subgroups and mixture of subgroups. Examples from oncology studies are used to illustrate the application of the proposed inference procedure.