Abstract: Supercritical carbon dioxide (SCO₂) is a kind of promising energy transfer and conversion working medium, which has been widely used in industrial field, such as electric power generation, supercritical fluid extraction, and seawater desalination. In recent years, a once-through Brayton cycle reactor is currently under R&D. It adopts CO₂ as the coolant which flows through the reactor fuel assembly to absorb the fission heat. There is no liquid-vapor transition beyond the critical pressure (7.38 MPa for CO₂), however, the thermophysical properties of CO₂ vary dramatically near the pseudo-critical temperature, which may lead to heat transfer deterioration. Besides, researches on heat transfer of SCO₂ at home and abroad focus on simple channels such as tubes. The effect of flow geometry on heat transfer has been seldom studied so far. In fact, channel geometry has certain influence by introducing potential effects, e.g., lateral mixing, uneven flow distribution, or flow pulsation. In the present study, a large eddy simulation (LES) was conducted to study the flow behavior and heat transfer characteristics of SCO₂ in tube, annular channel and sub-channel. The reason for choosing the three geometries lies in the fact that they are frequently selected as the representative flow geometry to prototypical fuel assembly. The numerical simulation was accomplished in OpenFOAM platform using WALE sub-grid scale model. A set of parameters were set identical for the three flow geometries before conducting LES study, which includes inlet Reynolds number, thermal equivalent diameter, Grashof number and non-dimensional heat flux. The flow geometry has an identical length of 45D, inlet Reynolds number of 13 028, inlet temperature of 301.15 K, inlet pressure of 8.0 MPa and wall heat flux of 36 kW/m2. The ability of LES method and WALE model in predicting heat transfer of SCO₂ was examined against the experimental data, showing acceptable agreement. The overprediction to the wall temperature may be attributed to the relatively large sub-grid scale Prandtl number adopted (default is 1.0 in OpenFOAM). Attention was paid to clarify the effects of flow geometry on the wall temperature, crosssection velocity distribution and turbulent statistics. Typical heat-transfer correlations were evaluated against the present LES results. The results show that, under the same conditions, the heat transfer performance in sub-channel is better than that of tube and annular channel, which is mainly due to the lateral flow and mixing of coolant in sub-channel. Among the six heat-transfer correlations selected, Dittus-Boelter correlation can not accurately predict the Nusselt number of SCO₂, while Kim correlation has a very high prediction accuracy for the heat transfer of SCO₂ in the three channels.