The paper presents an assessment of the performance of a variety of turbulence models in simulating buoyancy-aided, turbulent mixed convection in vertical pipes. This has been done by comparison of RANS predictions with DNS results already available in the literature. Both the RANS and the DNS studies were conducted for conditions of constant and uniform fluid properties with the influence of buoyancy being accounted for using the Boussinesq approximation. This eliminated effects of non-uniformity of properties other than through the action of buoyancy and enabled its influence to be considered in isolation. In the course of the study, the turbulence models have been classified into two groups, namely, those which were able to capture the main features of buoyancy-influenced heat transfer (Group one) and those that were not able to do so (Group two). Common features in model formulation have been identified for each group. It is shown that the response to buoyancy of commonly-used controlling parameters in turbulence damping functions varies significantly and that the performance of a model can largely be correlated with the type of controlling parameter used. A significant defect of the Group-one models which has been identified is that they continue to predict that the 'viscous sub-layer' remains thick as a result of the influence of buoyancy even when the velocity profile has been distorted to an extent that it has become inverted in the core, whereas DNS data clearly show that this is not the case. The use of different methodologies for modelling direct production of turbulence through the direct action of buoyancy has been shown to have little effect on predictions of mixed convection in vertical flows because the effect of buoyancy on turbulence is predominantly due to the indirect effect. (C) 2007 Elsevier Ltd. All rights reserved.