The high- and low-Ti basalts of the Parana-Etendeka province were primarily derived from old, trace element-enriched source regions in the lithospheric mantle, and they are associated with dyke swarms of different orientations. These swarms appear to reflect different amounts of extension, and it is inferred that the high- and low-Ti magma types were characterised by different melt generation rates of similar to 0.15 km(3) yr(-1) and similar to 0.4 km3 yr(-1),respectively [Stewart et al. Earth Planet. Sci. Lett. 143 (1996) 95-109]. There is probably a gap of similar to 2 Myr between the end of the main phase of CFB magmatism and the oldest rocks on the adjacent ocean floor. A simple numerical model has been used to constrain the amounts and rates of melt generated from the continental lithosphere and asthenosphere under finite duration extension. Melting in the mantle is assumed to be controlled by the dry peridotite solidus in the asthenosphere and the hydrous (0.2% H2O) peridotite solidus in the lithosphere. For a maximum beta of 4 and a duration of extension of 10 Myr, the derivation of melt from the asthenosphere by dry peridotite melting depends primarily on potential temperature (T-p) and is relatively insensitive to the thickness of the MEL, while the converse is the case for melt derived from the lithosphere by hydrous peridotite melting. For a T-p of 1450 +/- 50 degrees C inferred from the crustal thickness estimates along the Rio Grande Rise and Walvis Ridge, the model successfully generates 2-4 km of lithosphere-derived melt before producing significant volumes of asthenosphere-derived melt. It is concluded that increases of melt volume with time can be generated by decompression melting of the mantle lithosphere. Critically, in areas of significant melt generation within the mantle lithosphere during extension and break-up, there is likely to be a gap in the volcanic record between the end of melt generation in the lithosphere and the onset of melting in the underlying asthenosphere. No such gap is present if all melts are generated within the mantle plume, and thus these models may in principle be tested in the geologic record. (C) 2000 Elsevier Science B.V. All rights reserved.