Turbulent fluid flows with a swirl occur in numerous engineering practice situations. Five widely recognized turbulence models were evaluated using experimental results. Among the RANS-based two-equation models, the SST κ–ω model proved to be the most effective. The predictions obtained from a LES turbulence model were slightly better, however there was an enormous difference in the CPU time. The CPU time needed for the LES solution was 155.3 days, whereas it was only 14.2 days for the SST κ–ω solution. In this light, it is believed that the SST κ–ω model is the most efficient of those investigated. An important input for the numerical simulation of turbulent flow and heat transfer is the turbulence intensity at the inlet of a solution domain. For the evaluation of turbulence models, use was made of the measured turbulence quantities from the verification experiments. More often, simulations are executed based on the uniform values of the turbulence intensity across the inlet. The errors in the heat transfer results due to this practice are evaluated and are shown to be significant.