A numerical study of the evaluation of turbulence models for predicting the thermal stratification phenomenon is presented. The tested models are the elliptic blending turbulence model (EBM), the two-layer model, the shear stress transport model (SST), and the elliptic relaxation model (V2-f). These four turbulence models are applied to the prediction of a thermal stratification in an upper plenum of a liquid metal reactor experimented at the Japan Nuclear Cooperation (JNC). The EBM and V2-f models predict properly the steep gradient of the temperature at the interface of the cold and hot regions that is observed in the experimental data, and the EBM and V2-f models have the capability of predicting the temporal oscillation of the temperature. The two-layer and SST models predict the diffusive temperature gradient at the interface of a thermal stratification and fail to predict a temporal oscillation of the temperature. In general, the EBM predicts best the thermal stratification phenomenon in the upper plenum of the liquid metal reactor.