In designing a phase II cancer clinical trial monitoring simultaneously the response and toxicity rate of a therapeutic agent, the odds ratio has to be specified. The false positive or Type I error rate, however, can be substantially inflated if the specified value is considerably larger than the true odds ratio. To overcome the sensitivity of the error rates to the odds ratio, an adaptive procedure is proposed which allows the sample size to be re-estimated based on observed odds ratio. Simulation results show that the procedure is robust against the odds ratio assumptions and controls effectively the Type I error rate.