Traditionally, strict frequentist rules (e.g. two independent pivotal trials both with a significant result) have been developed to register a new medical treatment. On the other hand, there is a demand for flexibility to expedite the development process. A seamless (operational and/or inferential) phase II/III design usually contains two stages of traditional clinical trials, a learning stage (phase IIb) for selection and a confirmatory stage (phase III). The integration of two stages into a single trial accelerates the development process of new drugs and all accumulated data can be utilized in the decision process. In this presentation, a seamless inferential phase II/III design was demonstrated with adaptive treatment selection based on combination tests. An optimal sample size was determined with the familywise type I error rate controlled in the strong sense at a targeted power. Extensive simulations were conducted and operating characteristics were shown based on various dose-response models.