Our main constraint on the volume of melt coming out of the mantle at mid-ocean ridges is the thickness of oceanic crust determined by seismic methods. The recovery of serpentinized peridotites in the rift valleys of slow-spreading ridges has led recently to a revival of the suggestion that the seismically defined lower oceanic crust may include partially serpentinized peridotite. Both seismic and geochemical data from a recent experiment on the very slow-spreading Southwest Indian Ridge indicate that, away from the fracture zone valley, a total melt thickness of approximately 4 km is being generated. However, in one region, around Ocean Drilling Program (ODP) hole 735B, where approximately 2 km of the upper crust is estimated to have been removed, the seismic data indicate a crustal thickness of 5 km. Here, the seismically determined Moho is interpreted as a serpentinization front. Serpentinization is facilitated if there are readily available pathways for seawater to reach the upper mantle, and if the upper mantle is cool. These conditions frequently are met at fracture zones on slow-spreading ridges, in the ocean-continent transition zone at non-volcanic rifted margins, and at the axes of extinct rifts. In all these locations, serpentinization sufficient to lower mantle velocities to normal oceanic crustal velocities does not reach a depth of more than 5 km beneath the seabed, and commonly normal mantle velocities of approximately 8 km s (super -1) are also reached at shallow depths. Therefore, although in some anomalous locations the seismically defined crustal thickness can only be used as an upper limit on the melt supply, for normal oceanic crust, where the seismically defined crustal thickness is approximately 7 km, the Moho is probably not a serpentinization front, but rather a petrological boundary between mafic rocks above and ultramafic rocks below.