Computational simulations of heat transfer to fluids at a supercritical pressure have been performed using an ‘in-house’ CFD code written for two-dimensional axisymmetric flow and heat transfer based on the Favre averaging approach. Results are compared with recently available direct numerical simulations (DNS) which provide a benchmark dataset ideal for model assessment. The objective of the present study is to evaluate the performance of low-Reynolds number turbulence models in predicting mixed convection heat transfer to fluids at supercritical pressure, especially paying attention to the features which enable them to respond to the modifications of the turbulence field due to influences of flow acceleration and buoyancy. It has been found that a group of turbulence models which were previously found closely reproducing mixed convection under conditions of constant properties do not perform well for flows considered in the present study due to an over-response to changes in the flow. Models which were less successful previously perform better. The V2F model performs the best among all models tested. For strong-buoyancy-influenced cases, most models are able to reproduce turbulence recovery reasonably well but not the improvement on heat transfer. This is attributed, at least partly, to the inability of turbulence models in reproducing turbulent heat flux using a constant turbulent Prandtl number. The influence of the lack of a suitable description of the axial turbulent heat flux has been shown to be insignificant except immediately after the commencement of heating.