We present an adaptive model-based procedure for dose finding in phase I/II clinical trials when both efficacy and toxicity responses are available. In this setting, previous designs aimed at identifying the maximum tolerated dose as a surrogate for efficacy or the most successful dose, defined as the dose with the highest probability of efficacy without toxicity. Rather than using this definition of success, we propose considering all responses conditionally on the probability that dose-limiting toxicity is under a pre-specified threshold. The presented approach uses a joint model for the probability of an efficacy response and toxicity, and is evaluated through simulations. A retrospective application to a Phase I trial conducted in chronic lymphocytic leukemia is presented. Copyright © 2011 John Wiley & Sons, Ltd.