Summary.We study a stochastic control problem arising in the context of utility maximization under model uncertainty. The latter is formulated as a sup-inf problem over strategies π and models (measures) Q, and we treat the inner problem of minimizing over Q the sum of a Q-expected utility term and a penalty term based on the relative entropy of Q with respect to a reference measure P . We prove in general that there exists a unique optimal measure Q * and show that Q * is equivalent to P . For a continuous filtration, we characterize the dynamic value process of our stochastic control problem as the unique solution of a generalized backward stochastic differential equation with a quadratic driver. Our results extend earlier work in [21] and are based on a different approach.Mathematics Subject Classification (2000): 93E20, 91B16, 60H10, 46N10