Melting experiments have been performed on anhydrous mantle peridotite KLB-1 at 5 to 22.5 GPa using the multianvil press, with special attention paid to precision and accuracy of temperature and pressure measurement, oxygen fugacity, equilibrium, temperature gradient, and the effects of temperature gradient on the phase diagrams. The new phase diagram reveals complexities in the liquidus and solidus phase relations that were not reported by Takahashi (1986). At no pressure do the liquidus and solidus converge to a common temperature or to a narrow range of temperatures, a result that refutes the conjecture that mantle peridotite formed on the solidus as a partial melt (Herzberg and O'Hara, 1985) or a residual liquid. However, mantle peridotite could have formed as a cotectic liquid because KLB-1 exhibits nearly cotectic liquidus crystallization behavior in garnet plus magnesiow¿stite from 18 to 22 GPa. The implication is that mantle peridotite may be the product of a large-scale differentiation event and that MgO/SiO2 for mantle peridotite may have been phase equilibrium controlled by cotectic crystallization in a magma ocean. Fractionation of magnesiow¿stite or majorite garnet is inferred at 18 to 22 GPa for a bulk Earth MgO/SiO2 that is higher or lower than mantle peridotite, respectively. Alternatively, if the Earth formed from materials that were the same in MgO/SiO2 as mantle peridotite, then the liquidus phase equilibria reported here may be irrelevant or incidental in understanding Earth structure. ¿ American Geophysical Union 1994