The compressibility of basic melt at 1 atmosphere is about an order of magnitude higher than that of mantle minerals. Consequently, the density contrast between melt and the principal residual crystals in mantle source regions is expected to decrease with increasing source region depth. The increasingly olivine-normative character of primary melts produced at greater depths is also expected to result in a decrease in this density contrast with increasing source region depth. Once vertical permeability is established by melt generated during partial melting, buoyancy-driven melt percolation can under some circumstances segregate melt from the residual crystals in its source region on a geologically rapid time scale. Limits to this process are provided by cooling of the source region (freezing melt in) and rigidity of the crystalline matrix (mechanically trapping melt). Source region size influences these limits strongly: consequently, small, partially molten diapirs (~km in diameter) may be able to trap large melt fractions (>30%), but larger source regions would be unable to do so. The reduction in density contrast with pressure reduces the buoyant force driving melt percolaton and provides another limit to melt segregation. Diapirs at depth may thus stably contain large fractions of melt but may decompress and unload their melt during ascent; this effect would be enhanced in small diapirs and may be relevant to the genesis of komatiitic magma. Melt compression may also be a factor in explaining why the very different maximum depths inferred for typical basic melt segregation from source regions on different planets- ~500 km on the moon, ~250 km on Mars, ~100 km on earth-correspond to similar pressures (2535 kbar); at greater pressures, melt may no longer be capable under ordinary conditions of segregating upwards by buoyancy. This may also help to explain why depleted peridotites overlie more fertile peridotites and how deep regions of the mantle are able to remain fertile over geologic time.