The accurate prediction of the flow field in a long reach of a complex waterway is prerequisite to the evaluation to acceptable engineering accuracy of strategies for the restoration, resource management and hazard assessment. In this study, we consider some aspects of the numerical prediction of flow in a long reach of the Sacramento River in California with view to gaining a better understanding the sources and the extent of the uncertainties that are present in the predictions. The predictions were obtained using a finite-volume method to solve the time-averaged forms of the equations governing the conservation of mass and momentum. Body-fitted coordinates were used to define the complex river bathymetry. Both depth-averaged and fully three-dimensional (3D) computations were performed. The results were compared with data obtained from a large-scale physical model of the same reach. The outcome of the comparisons served to provide the basis for the assessment of the sensitivity of the solutions to assumptions made regarding the conditions at inlet to the solution domain, the choice of turbulence model used to close the time-averaged equations and, for the depth-averaged calculations, the choice of method used for prescribing the bed friction. The influence of other parameters such as the density of the computational grid, the method used to define the channel bathymetry, and the unsteadiness of the mean flow that results from vortex shedding from a bridge pier and abutments were also assessed. The results obtained form a useful addition to the growing literature on the computational modelling of flows in natural waterways.