Low-Reynolds-number k-epsilon turbulence models have been successfully used by numerous researchers in various applications. It has been found that the Myong-Kasagi model (MK) among them outperforms in simulations of thermal-fluid fields at supercritical pressures and near the corresponding pseudo critical temperature. However, they are used as is without a clear understanding of the cause of the good performance. In this paper, several well-known low-Reynolds-number turbulence models, including MK, are critically reviewed against DNS data and RANS calculation results to find the reasons, if any exist, for the superiority of MK model. The most outstanding factor identified may be the fact that MK introduced the Taylor microscale as the near-wall length scale and combined it with the integral length to result in a combined turbulence length scale, which is valid over the entire range of a turbulent boundary layer. The eddy viscosity formula with the incorporation of the turbulence length scale is naturally expected to provide a better representation of flows with strong buoyancy due to wall heating, especially in the near-wall region, where the buoyancy effect mainly occurs. As a result, MK-simulated highly buoyant flows showed excellent agreement with experimental data when applied with the property-dependent turbulent Prandtl number and shear-stress-dependent damping length. A comparison with DNS data of the turbulence data obtained from RANS calculations with MK also showed a good agreement. (C) 2016 Elsevier Ltd. All rights reserved.