The Westinghouse Electric Company, together with an international team, is developing its next-generation high-capacity nuclear power plant based on lead-cooled fast reactor technology. In general, the Prandtl number of a heavy metal fluid is two orders of magnitude lower than that of water. Because of the low Prandtl number of lead, the thermal boundary layer in the temperature field is much thicker than the viscous boundary layer in the velocity field. Existing engineering turbulence models all use the Reynolds analogy for coupling temperature and velocity fields, but it is not valid for Heavy Lead Metal flow. Therefore, the selection of an appropriate turbulent Prandtl number (Prt) is crucial for lead fluid simulation. The conventional choice of Prt as a constant value of 0.9 is not valid for low-Prandtl-number fluid flow. This paper aims at using an advanced high-fidelity numerical approach, Large Eddy Simulation (LES), to provide detailed insight into the physics of the flow and the associated heat transfer in a rod bundle with liquid metal flows. The results for the global Nusselt number with different Prt values (0.9 ∼ 3.0) indicate that a higher Prt can reduce the predicted Nusselt number by up to 44.1 %. The detailed temperature distributions obtained with RANS and LES are compared to better understand the deviation introduced by the turbulence model. For turbulent flow, the correlations for Nusselt number proposed by Cheng and Tak (2006) are in good agreement with the simulation, especially for a Peclet number < 1000. The analysis shows that the RANS model with Prt = 1.5 shows better agreement on the prediction of local temperature distribution and global Nusselt number. Although the modified local turbulent Prandtl number model shows some improvement over the constant Prt model, the development of a more sophisticated local turbulent Prandtl number model is necessary for further applications.