To assess the accuracy of large eddy simulation (LES) predictions for a flow in a rod bundle, analyses were performed with different parameters of a constant-coefficient Smagorinsky LES model for a flow in a square-pitch rod bundle, and model predictions are compared with experimental data. The parameters considered are the grid structure, the value of the Smagorinsky constant, the damping of the eddy viscosity, and the size of the channel geometry. Because LES simulations are computationally very demanding, for adequately accurate predictions the grid structure needs to be well optimized in terms of cell size, aspect ratio, and cell orthogonality. The use of hanging nodes can significantly reduce the number of cells without a significant penalty on the accuracy of predictions. For this flow, the change in the value of the Smagorinsky constant from 0.14 to zero did not have a drastic effect on predictions. Although, overall, Lilly damping gave slightly better predictions than van Driest damping, both damping functions gave similar predictions. The LES predictions for the mean axial velocity, for the fluctuating velocity component in the main flow direction, and for the Reynolds stresses are in very good agreement with the experimental measurements. There is also good agreement between predictions and measurements for the wall shear stress, but there is a significant discrepancy between predictions and measurements for the fluctuating velocity components in the lateral directions (u and v).