Computational fluid dynamics is employed for detailed prediction of the hydrokinetic turbine performance and wake modelling. Of these, Reynolds-averaged Navier-Stokes (RANS) models are most widely used due to their ability to resolve power performance and detailed flow features at relatively low computational costs and acceptable accuracy. The limitations of these models are often not well understood when applied to complex turbine and wake dynamics which could lead to potential inaccurate and inappropriate conclusions. This paper focuses on the prediction of the wake generation, dissipation and flow recovery using commercially available modelling software. The approach and findings of previous numerical investigations on this matter are reviewed and compared to experimental measurements reported for a dual-rotor reference turbine. The shortcomings of these models are discussed and appropriate modelling techniques for the preliminary design or analysis of hydrokinetic turbines and inland energy generation schemes are identified. Commercially available RANS models show a good correlation of turbine performance. However, prediction of the wake behaviour is improved by using a virtual disk model with the blade element momentum theory, employing Reynolds stress closure models. These models allow for modelling the anisotropic conditions in the wake unlike the more popular eddy viscosity models. In addition, simplified rotor geometry models using blade element momentum theory are found to adequately model wake development and dissipation at a modest computational expense. The shortcomings of other approaches in terms of wake dissipation prediction and the effect of boundary and inflow conditions are analysed, emphasizing the importance of correct prescriptions of model parameter.