Knowledge of the thermal structure and processes in subduction zones is so uncertain that our more extensive knowledge of source materials and phase relationships remains insufficient for prediction of magmatic products. The phase relationships do provide important constraints for testing geophysical and petrogenetic models, especially when considered along with geochemical constraints. Physical questions, such as the behavior of partially molten rocks, are intimately related to the compositions of magmas from source to eruption site. Tentative predictions and conclusions include: (1) Andesite is not a primary magma from oceanic crust. (2) It is unlikely that liquids from subducted oceanic crust yield andesites by fractionation, except by melting of amphibolite in an especially warm subducted crust. (3) It is more likely that hydrous siliceous magmas from the crust leak into the overlying mantle. (4) Andesite magmas can be generated in peridotite only under exceptional circumstances. (5) H2O-undersaturated basic magmas from peridotite modified by aqueous fluids or hydrous magma may yield andesite by fractionation. (6) Phlogopite peridotite could escape melting and yield alkalic magmas at deeper levels. (7) Andesite magmas could be produced in continental crust only by extreme heating through underplating of basalts or by mixing of basalt and rhyolite (derived from crust). Most of these conclusions need revision if convection in the asthenosphere wedge raises the temperature of the subducted slab to 1250°C at 100 km as proposed by Marsh (1979a); this demonstrates the dependence of experimental predictions on geophysical models.