This paper treats the practical and challenging control problem of tracking a prescribed continuous trajectory for an autonomous underwater vehicle immersed in the presence of gravity, buoyancy, hydrodynamic and other uncertain forces and moments. These uncertain forces and moments are bounded and may be difficult to model accurately, but they act persistently or over long periods of time. The trajectory is specified in terms of desired attitude and translational motion for a rigid body model of the vehicle. For an autonomous underwater vehicle (AUV) in a dynamic environment like the ocean, the presence of changing ocean currents create hydrodynamic forces and moments that are not well-known or predictable, even though they are bounded. In the absence of such forces and moments, it is possible to model the AUV dynamics very accurately and use the model for trajectory generation and tracking, as has been shown in prior research. However, the presence of uncertainties makes it difficult to accurately model the dynamics. This in turn makes the control task of tracking a desired or prescribed trajectory very challenging as an accurate model of the system dynamics is not available. We develop a robust feedback tracking scheme for autonomous underwater vehicles that can track a prescribed trajectory, while rejecting the effects of disturbances due to poorly known inputs with a simple yet general internal model.