We present a new thermal model that resolves the discrepancy between observations of narrow (~4 km) magma chambers on the northern East Pacific Rise (EPR) and predictions based on previous thermal models of wide (~20 km) chambers by including cooling of the lower crust by convection within the magma chamber. For this cooling to produce a solid crust within a few kilometers of the axis the temperature of the magma chamber must be significantly below the solidus temperature of the lower crust. This model is internally consistent for an average temperature of 1240¿C for primary magma leaving the mantle and an average magma changer temperature of 1150¿C. These temperatures imply ~10% MgO for primary magma, which is at the low-Mg end of the range of proposed compositions, and ~6% MgO for magma residing in the changer which is comparable to the most evolved basalts commonly erupted along the EPR. There is a trade-off in the inputs to the model such that raising the primary magma temperature requires a lower magma chamber temperature to maintain the same maximum temperature of the lower crust. It is reasonable for the magma chamber to be more evolved than most erupted lavas if magma mixing exerts substantial control over the composition of the erupted lavas. Aphyric 10% MgO basalt should be buoyant in 6% MgO basalt, therefore the mixing need not involve the entire chamber prior to eruption. ¿ American Geophysical Union 1988