Abstract: Based on two sets of experimental data, the performance of predicting heat transfer of supercritical water in a vertical tube using CFX code with 12 turbulence models was assessed. The result of BSL explicit algebraic Reynolds stress model agrees the most closely with the experimental data under forced convection, but the differences among various turbulence models are small. However, under mixed convection condition, ε type turbulence models with wall function can not predict heat transfer deterioration phenomenon, while ω type turbulence models with automatic wall treatment can, but the predicted wall temperatures deviate largely from experimental results. The assessment results show that near wall treatment methods have great influence on the simulation results. The following two factors also lead to the unreasonably predicted heat transfer behavior: Modeling turbulent heat flux with turbulent Prandtl number and neglecting density fluctuation. It is recommended that turbulence models should be improved.