An in-depth understanding of the liquid metal flow and heat transfer is essential in order to identify the key mechanisms for the hearth erosion of a blast furnace. In this study, a comprehensive computational fluid dynamics model is described which predicts the flow and temperature distributions of liquid iron in blast furnace hearth, and the temperature distribution in the refractories. The new model addresses conjugate heat transfer, natural convection and turbulent flow through porous media, with its main features including improved transport equations (a modified k-e turbulence model and thermal dispersion term) and a three-dimensional, high-resolution grid. The new turbulence model and terms take account of the effect of microscopic flows around coke particles and allow unified treatment of coke bed and coke-free layer. The predicted results show a well-organized flow pattern: two large-scale recirculation zones are separated vertically at the taphole level. This flow pattern controls the temperature distribution in the liquid phase, so that the temperature remains nearly uniform in the upper zone, but changes mainly across the lower zone. The effects of several factors were examined, such as cases comparing fluid buoyancy with constant fluid density as well as the shape and position of the coke free zone (i.e. based on reported dissection studies). Natural convection is found to be most important for the liquid metal flow patterns observed. Comparison with the plant data shows that the refractory pad temperature is under-predicted when assuming intact hearth lining. The pad temperature is very sensitive to the erosion of protection layer in the hearth lining.