The crossflow is the key phenomenon in turbulent flow inside rod bundles. In order to establish confidence on application of computational fluid dynamics (CFD) to simulate the crossflow in rod bundles, three Reynolds-Averaged Navier Stokes (RANS) models i.e. the realizable k-ε model, the k-ω SST model and the Reynolds stress model (RSM), and the Large Eddy simulations (LES) with the Wall-Adapting Local Eddy-viscosity (WALE) model are validated based on the Particle Image Velocimetry (PIV) flow measurement experiment in a 5 × 5 rod bundle. In order to investigate effects of periodic boundary condition in the gap, the numerical results obtained with four inner subchannels are compared with that obtained with the whole 5 × 5 rod bundle. The results show that periodic boundaries in the gaps produce strong errors far downstream of the spacer grid, and therefore the full 5 × 5 rod bundle should be simulated. Furthermore, it can be concluded, that the realizable k-ε model can only provide reasonable results very close to the spacer grid, while the other investigated models are in good agreement with the experimental data in the whole downstream flow in the rod bundle. The LES approach shows superiority to the RANS models.