Benchmark experiments simulating flows in a pressurized water reactor rod bundle were analyzed to evaluate the performance of a state-of-the-art computational fluid dynamics (CFD) code. For the simulation of turbulence a number of standard k-[curly epsilon] models were used. Away from components that cause significant flow deflections, the difference between mean velocity predictions and measurements is within the experimental error. Near such components there is significant discrepancy between velocity predictions and measurements. Even in rod bundles without flow deflectors, the turbulence predictions of standard k-[curly epsilon] models show significant discrepancy with measurements. These discrepancies are greater near components that cause flow deflections. Turbulence generated by vanes on spacer grids significantly enhances thermal mixing. To improve the fidelity of CFD simulations of flows in reactor rod bundles, the development of Reynolds averaging of the Navier-Stokes equations turbulence models based on such flows is needed.