In this paper the performance of various turbulence models potentially suitable for the prediction of indoor air flow and temperature distributions in an atrium space was evaluated in a systematic way. The investigation tested these models for various thermal conditions in atria of different geometrical configuration in two existing buildings using the Reynolds Averaged Navier-Stokes (RANS) modeling approach. The RANS turbulence models that were rested include the one-equation model (the Spallart-Allamaras) and two-equation models (the standard k-epsilon, RNG k-epsilon, realizable k-epsilon, standard k-omega and SST k-omega models). The radiation exchange between the surfaces of the atrium space was considered using the Discrete Transfer Radiation Model (DTRM). The resultant steady state governing equations were solved using a commercial CFD solver FLUENT. The numerical results obtained for a particular time of the day were compared with the experimental data available. Relatively good agreement between the experimental and CFD predictions was obtained for each model. However, from the results obtained, it was found that the performance of two-equation turbulence models is better than one-equation model and among the two-equation models, the SST k-omega model showed relatively better prediction capability of the indoor environment in an atrium space than k-epsilon-models. (C) 2012 Elsevier B.V. All rights reserved.