The relationship between the abundances of Th, U and Pb isotopes in basalt melts and the [Th/U] ratio of their source is assessed. A simple melting model is used to show that whereas the activity ratio ( 230 Th/ 232 Th) in the initial melt before extraction is equal to the bulk source ratio, that in the extracted melt may be higher. The difference depends upon the rate of melting relative to the half life of 230 Th (73 ka). Only when the rate is fast compared to this half life will ( 230 Th/ 232 Th) in the extracted melt provide a correct estimate of [ 232 Th/ 238 U] in the source and therefore of its [Th/U ] ratio. This is normally not the case for MORB, and a better estimate of source [Th/U ] ratio is derived from [ 232 Th/238U] ratio in the basalt, which does not depend upon the rate of melting. Available d a ta for MORB glasses give a best estimate for their source [Th/U] = 2.58 ± 0.06. This value is less than both that of the bulk Earth of 3.9 ± 0.1, and of the source of plume basalts from Iceland and Hawaii, which are 3.3 and 3.2 respectively. These estimates contrast with the [ 232 Th/ 238 U] ratio required to produce the radiogenic { 208 Pb/ 206 Pb} atomic ratio of MORB over 4.55 Ga. This averages 3.8 and is little different from the average derived from Pb-isotopes in plume basalts. These observations are most easily reconciled if Th, U and Pb are efficiently stripped from the mantle by melting and have a residence time there of ≤1 Ga. The [Th/U ] ratio of 2.6 for the upper mantle requires melt fractions of ≤1 % to be involved in transferring U and Th from this region into the continents. Such melt fractions are present in subduction zones and in the source regions of continental alkali basalts.