Re and Al 2 O 3 were heated with laser beams from both sides. Acting like planar heat sources, the two`hot plates' eliminate the axial temperature gradient in the sample between the plates. Temperature variation is less than 3% within roughly 30 mm diameter at 2,500 K. Before the melting experiments, the sample was scanned with a laser beam and heated to about 2,000 K to reduce the pressure gradient and to produce a high-pressure solid-phase assemblage. For stable and smooth temperature control, temperatures were increased by adjusting an aperture placed near the beam exit, stepwise, instead of by adjusting power. Each step corresponds to a 50±100 K increase. A 30-mm spot was homogeneously heated by opening the aperture (increasing the step). At the onset of melting, temperature remains constant or drops slightly with the step increment, and then drastically increases (.400 K) within one step. To ensure the reliability of the melting criteria used in this study, we conducted melting experiments at pressures (16±27 GPa) overlapped by the multi-anvil apparatus and the diamond-anvil cell, using the same starting material, and obtained consistent melting temperatures (Fig. 3). We also used the same melting criteria to determine the melting temperature of MgSiO 3 ±perovskite previously studied by other investigators, and our results agree with these recent determinations 13,14 (Fig. 3). The temperature runaway phenomena near the onset of melting observed in simple and complex samples were probably a result of the latent heat of melting, followed by melt migrating away from the heated spot because of the large thermal pressure and, ®nally, the Re foils would have been heated without sample in between. No chemical reaction between Re and sample was observed in the multi-anvil experiments on a scale of 1 mm. The melting temperatures reported here are the last temperatures before melting sets in. Pressures were measured using a ruby-¯uorescence technique after each measurement of melting temperature.