Piston-cylinder experiments on a Pleistocene adakite from Mindanao in the Philippines have been used to establish near-liquidus and sub-liquidus phase relationships relevant to conditions in the East Philippines subduction zone. The experimental starting material belongs to a consanguineous suite of adakitic andesites. Experiments were conducted at pressures from 0.5 to 2 GPa and temperatures from 950 to 1,150°C. With 5 wt. % of dissolved H2O in the starting mix, garnet, clinopyroxene and orthopyroxene are liquidus phases at pressures above 1.5 GPa, whereas clinopyroxene and orthopyroxene are liquidus (or near-liquidus) phases at pressures <1.5 GPa. Although amphibole is not a liquidus phase under any of the conditions examined, it is stable under sub-liquidus conditions at temperature ≤1,050°C and pressures up to 1.5 GPa. When combined with petrographic observations and bulk rock chemical data for the Mindanao adakites, these findings are consistent with polybaric fractionation that initially involved garnet (at pressures >1.5 GPa) and subsequently involved the lower pressure fractionation of amphibole, plagioclase and subordinate clinopyroxene. Thus, the distinctive Y and HREE depletions of the andesitic adakites (which distinguish them from associated non-adakitic andesites) must be established relatively early in the fractionation process. Our experiments show that this early fractionation must have occurred at pressures >1.5 GPa and, thus, deeper than the Mindanao Moho. Published thermal models of the Philippine Sea Plate preclude a direct origin by melting of the subducting ocean crust. Thus, our results favour a model whereby basaltic arc melt underwent high-pressure crystal fractionation while stalled beneath immature arc lithosphere. This produced residual magma of adakitic character which underwent further fractionation at relatively low (i. e. crustal) pressures before being erupted.