The subduction of oceanic lithosphere into the Earth's deep interior is thought to drive convection and create chemical heterogeneity in the mantle. The oceanic lithosphere as a whole, however, might not subduct uniformly: the fate of basaltic crust may differ from that of the underlying peridotite layer because of differences in chemistry, density and melting temperature. It has been suggested that subducted basaltic crust may in fact become buoyant at the mantle's 660-km discontinuity, remaining buoyant to depths of at least 800 km, and therefore might be gravitationally trapped at this boundary to form a garnetite layer(1,2). Here we report the phase relations and melting temperatures of natural mid-ocean ridge basalt at pressures up to 64 GPa (corresponding to similar to 1,500 km depth). We fmd that the former basaltic crust is no longer buoyant when it transforms to a perovskitite lithology at about 720 km depth, and that this transition boundary has a positive pressure-temperature slope, in contrast to the negative slope of the transition boundary in peridotite. We therefore predict that basaltic crust with perovskitite lithology would gravitationally sink into the deep mantle. Our melting data suggest that, at the base of the lower mantle, the former basaltic crust would be partially molten if temperatures there were to exceed 4,000 K.