The position of the transition from spinel lherzolite to garnet lherzolite in the system CaO-MgO-Al2O3-SiO2 (CMAS) has been determined experimentally at near-solidus temperatures. In reversed experiments, the transition occurs between 18 and 20 kbar at 1200 °C and between 26 and 27 kbar at 1500 °C, corresponding to higher pressures than previously envisaged. A position for the transition deeper within the Earth further complicates the explanation of the so-called garnet signatures in the trace element and isotope patterns of mid-ocean ridge basalts. If melting during adiabatic upwelling beneath a mid-ocean ridge begins at the depth required for the stability of garnet in peridotitic compositions, simple melting models predict that the amount of melt produced should be much greater than the observed thickness of the oceanic crust. A partial solution to the apparent conflict might be that (1) the rather simplistic melting models are in error, (2) that melting begins in garnet pyroxenite veins that are believed to be stable at lower pressures than garnet lherzolite or (3) that melting does not involve garnet at all, but it is clinopyroxene causing the trace element patterns observed in basalts erupted at mid-ocean ridges. A second set of reversal experiments were conducted to investigate the solubility of alumina in both orthopyroxenes and clinopyroxenes at the high temperatures near the solidus in the system CMAS. The results are compatible with most previous studies, and may be used as a starting point to calibrate thermodynamic models for pyroxenes in chemical systems, approximating upper mantle chemistry.