Geophysical evidence precludes the existence of a large, mainly molten magma chamber beneath portions of the East Pacific Rise (EPR). A reasonable model, consistent with these data, involves a thin (tens to hundreds of meters high), narrow (<1–2 km wide) melt lens overlying a zone of crystal mush that is in turn surrounded by a transition zone of mostly solidified crust with isolated pockets of magma. Evidence from the superfast spreading portion of the EPR suggests that the composition of the melt lens is mainly moderately fractionated ferrobasalt. These results have important implications for magmatic processes occurring beneath mid-ocean ridges and are consistent with a model that effectively separates the processes of magma mixing and fractionation into different parts of a composite magma chamber. Magma mixing, as evidenced by disequilibrium relations between host liquids and included phenocrysts, is especially apparent in samples from low magma supply ridges and probably mainly arises from interactions between crystals of the mush zone and new injections of primitive magma rising out of the mantle. Magmatic differentiation beneath mid-ocean ridges occurs in two parts. Migration of melts through the transition and mush zones can produce chemical trends consistent with in situ fractionation processes. Segregation of melt into molten horizons near the top of a composite magma chamber promotes the more extensive differentiation characteristic of fast spreading ridges. The optimum conditions for the formation of highly differentiated abyssal lavas is where small, discontinuous melt lenses occur, such as at intermediate spreading rates, in the vicinity of propagating rifts, and near ridge offsets at fast spreading ridges. Along-axis homogenization of subaxial magma is inhibited by the thin, high aspect ratio of the melt lens and by the high viscosities expected in the mush and transition zones. Low magma supply ridges are unlikely to be underlain by eruptable magma in a steady state sense, and eruptions at slow spreading ridges are likely to be closely coupled in time to injection events of new magma from the mantle. Extensional events at high magma supply ridges, which are more likely to be underlain by significant volumes of low-viscosity melt, can produce eruptions without requiring associated injection events. The critical magma supply necessary for the development of a melt lens near the top of a composite magma chamber is similar to that of normal ridges spreading at rates of about 50–70 mm/yr, a rate approximately corresponding to that marking an abrupt change in the morphology and gravity signal at the ridge axis. A composite magma chamber model can explain several previous enigmas concerning mid-ocean ridge basalts, including why slow spreading ridges dominantly erupt a narrow range of relatively undifferentiated lavas, why magma mixing is most evident in lavas erupted from slow spreading ridges, why fast spreading ridges erupt a wide range of generally more differentiated compositions, why bimodal lava populations occur in the vicinity of some propagating rifts, and how along-axis geochemical segmentation can occur at a scale shorter than the major tectonic segmentation of ridge axes.