There is a need in the IT industry for CFD models that are capable of accurately predicting the thermal distributions in high power density open-aisle air-cooled data centers for use in the design of these facilities with reduced cooling needs. A recent detailed evaluation of a small data center cell equipped with one high power rack using current CFD practice showed that the CFD results were not accurate. The simulation results exhibited pronounced hot/cold spots in the data center while the test data were much more diffused, indicating that the CFD model under-predicted the mixing process between the cold tile flow and the hot rack exhaust flow with the warm room air. In this study, a parametric study was carried out to identify CFD modeling issues that contributed to this error. Through a combined experimental and computational investigation, it was found that the boundary condition imposed at the perforated surfaces (e.g., perforated tiles and rack exhaust door) as fully open surfaces was the main source of error. This method enforces the correct mass flux but the initial jet momentum is under-specified. A momentum source model proposed for these perforated surfaces is found to improve the CFD results significantly. Another CFD modeling refinement shown to improve CFD predictions is the inclusion of some large-scale geometrical features of the perforated surfaces (e.g., lands/gaps) in the CFD model, but this refinement requires the use of grids finer than those typically used in practice.