Abstract: There might be a three-layer molten pool configuration with a heavy metallic layer at the bottom in the stratified molten pool configuration. Because of the decay heat in the heavy metallic layer and the relatively low critical heat flux (CHF) at the bottom of the molten pool, vessel failure could occur at the sidewall adjacent to the heavy metallic layer. The heavy metallic layer experiment was carried out using Wood’s metal or water as the simulant to analyze the heat transfer of the heavy metallic layer under different boundary conditions. The experimental apparatus mainly contains the cooling water tank, pump, valve, test section, and watercooling heat exchanger. The different cooling rates could be obtained by adjusting the valves. Wood’s metal test has a top cooling and a sidewall cooling boundary conditions, and water tests have a top heated and a sidewall cooling boundary conditions. The sidewall cooling channels were used to simulate the external reactor vessel cooling (ERVC) conditions. A three-dimension hemisphere with a radius of 1.2 m was used as the test section, and the heavy metallic layer’s height was set to 0.3 m. Heating coils were installed in the test section to simulate the decay heat in the heavy metallic layer. Besides, thermocouples were inserted in the sidewall and test section to record the temperature data. The experimental results illustrate that the melt temperature in the heavy metallic layer increases with the height. Thermal stratification is evident in the heavy metallic layer, and heat convection is limited. But the normalized melt temperature along the horizontal direction tends to be constant. When no crust formation occurs, the water test results demonstrate that the sideward heat flux increases with the polar angle. However, when the crust formation occurs in the heavy metallic layer, Wood’s metal test results demonstrate that the dominant heat transfer mode would turn from the convective heat transfer to the heat conduction. It will affect the sideward heat flux distribution that is not constant for both heat transfer modes. Consequently, the sideward heat flux would decrease with the increase of the polar angle and the maximum sideward heat flux occurs in the bottom of the test section. By comparing the experimental sideward Nusselt number with the calculated results of previous correlations, it is found that the heat transfer correlations of ACOPO, BALI, Mayinger, and UCLA could reasonably predict the average sideward Nusselt number of the heavy metallic layer when Wood’s metal or water is the simulant.