THE present high gradient of density in the upper mantle provides neutral buoyancy zones in which certain minerals may have been accumulated in the past or to which they may still be migrating. The necessary condition for progressive accumulation in such a zone is that the mineral concerned has opportunities of rising or sinking relative to its surroundings with a velocity which, averaged over long periods, is great enough to overcome the bulk convective drift if such occurs. Frank1 has pointed out the high heat transfer capability of a small volume fraction of melt fluid percolating through the bulk, compared with bulk convection. With percolation active, it is likely that a fluctuating distribution of migration and local concentration of fluid has been the dominating pattern of events in the mantle, at least down to 500 km, during a large fraction of the Earth's history. Hence it is unlikely that any portion of the mantle above this level has escaped epochs during which fluid has percolated through it. If during such an epoch of involvement in local fluidity the fluid fraction becomes sufficient to lower the effective viscosity to within an order of magnitude of that of a normal homogeneous melt, then insoluble crystals suspended in the partial melt, rising or falling under buoyancy forces, could well attain velocities a million-fold greater than any plausible bulk convection velocity of the mantle in which the localized partial melt is embedded.