Accurate design of the secondary air system is one of the main tasks for reliability and performance of gas turbine engines. The selection of a suitable turbulence model for the study of rotor-stator cavity flows, which remains an open issue in the literature, is here addressed over a wide range of operating conditions. A numerical benchmark of turbulence models is indeed proposed in the case of rotor-stator disk flows with and without superimposed throughflow. The predictions obtained by the means of several two equation turbulence models available within the CFD solver Ansys CFX 12.0 are compared with those previously evaluated by Poncet et al. through the Reynolds Stress Model (RSM) of Elena and Schiestel implemented in a proprietary finite volume code. The standard k-eps and k-w SST models including high and low Reynolds approaches, have been used for all calculations presented here. Further more, some tests were conduced using the innovative k-w SST-CC and k-w SST-RM models that take into account the curvature effects via the Spalart-Shur correction term and the reattachement modification proposed by Menter respectively. The numerical calculations have been compared to extensive velocity and pressure measurements performed on the test rig of the IRPHE's laboratory in Marseilles. Several configurations, covering a wide range of real engine operating conditions, were considered. The influence of the typical non dimensional flow parameters (Reynolds number and flowrate coefficient) on the flow structure is studied in detail. In the case of an enclosed cavity, the flow exhibits a Batchelor-like structure with two turbulent boundary layers separated by a laminar rotating core. When an inward axial throughflow is superimposed, the flow remains of Batchelor type with a core rotating faster than the disk because of conservation of the angular momentum. In this case, turbulence intensities are mainly confined close to the stator. Turbulence models based on a low Reynolds approach provide better overall results for the mean and turbulent fields especially within the very thin boundary layers. The standard k-w SST model offers the best trade-off between accuracy and computational cost for the parameters considered here. In the case of an outward throughflow, the k-w SST in conjunction with a low Reynolds approach and RSM models provide similar results and predict quite well the transition from the Batchelor to the Stewartson structures.