Abstract: In order to evaluate the performance prediction accuracy of axial flow pump under different turbulence models, the axial flow model used in South-to-North Water Diversion Project was study object, the standard k-ε、RNG k-ε(renormalization group k-ε)and RSM(reynolds stress model) are used to predict the performance of axial flow pump and numerically simulate its flow field based on SIMPLE(semi-implicit method for pressure-linked equations) algorithm and structured hexahedral mesh. The simulated head and efficiency error compared with experimental results which are tested by ministry of water resources in Tianjin were analyzed. The results showed that simulation results were affected by the mesh density, and the head and efficiency prediction errors were increased by the sparse grid in the large flow rata and small flow rata conditions, which would reach 3% or more; Predictive errors of the head for Standard k-ε, RNG k-ε and RSM turbulence model were 0.97%, 1.12% and 1.24% respectively, and the efficiency of errors were 2.93%, 2.49% and 2.97% under the optimal conditions, which could meet the needs of engineering applications. However, the flow field is complex because the secondary flow and cavitations exist under off-design conditions. The head maximum predictive error range of three turbulence models was 9.40%-14.30%, the efficiency was 4.48%-8.30%. The conclusions in this paper will provide a reliable performance prediction data and practice for the axial flow pump.