With the recent advances in computer technology, computational fluid dynamics (CFDs) has become an important tool to design and improve blood-contacting artificial organs, and to study the device-induced blood damage. Commercial CFD software packages are readily available, and multiple CFD models are provided by CFD software developers. However, the best approach of using CFD effectively to characterize fluid flow and to predict blood damage in these medical devices remains debatable. This study aimed to compare these CFD models and provide useful information on the accuracy of each model in modeling blood flow in circulatory assist devices. The laminar and five turbulence models (Spalart-Allmaras, k-ε (k-epsilon), k-ω (k-omega), SST [Menter's Shear Stress Transport], and Reynolds Stress) were implemented to predict blood flow in a clinically used circulatory assist device, the CentriMag centrifugal blood pump. In parallel, a transparent replica of the CentriMag pump was constructed and selected views of the flow fields were measured with digital particle image velocimetry (DPIV). CFD results were compared with the DPIV experimental results. Compared with the experiment, all the selected CFD models predicted the flow pattern fairly well except the area of the outlet. However, quantitatively, the laminar model results were the most deviated from the experimental data. On the other hand, k-ε renormalization group theory models and Reynolds Stress model are the most accurate. In conclusion, for the circulatory assist devices, turbulence models provide more accurate results than the laminar model. Among the selected turbulence models, k-ε and Reynolds Stress Method models are recommended.