Two of the most important sources of information on the style of mantle convection come from geochemical constraints and mantle tomography. Constraints imposed by uranium-thorium systematics are examined in this paper. Observations require that the mid-ocean ridge basalt (MORB) source region be nearly homogeneous with a thorium-uranium ratio &kgr;=2.5¿0.1. Simple mass balance considerations utilizing limits on the value of &kgr; for the continental crust (&kgr;<6) and a bulk earth value &kgr;=4.0¿0.2 with heat production constraints show that the whole mantle has a thorium-uranium ratio greater than &kgr;=3.35¿0.1. This leads to the conclusion that a substantial fraction of the mantle has a much more primitive value of &kgr; (closer to 4.0) than the MORB source region. Our model assumes a near-uniform, depleted upper mantle that is the source region for MORB and has &kgr;=2.5 and a near-primitive lower mantle with &kgr;=4.0. Our results indicate that it is very difficult to have the upper mantle reservoir larger than 65% of the whole mantle and our preferred value is 45%. Three alternative hypotheses can be proposed: (1) A vertical stratification of &kgr; through the mantle (since the role of whole mantle convection is to extract heat, the resulting vertical transport in the mantle would be expected to eliminate the vertical stratification of &kgr;), (2) a mantle permeated with blobs (since the blobs would have to constitute some 65¿10% of the mantle and be absent from the MORB source region, it is difficult to support this hypothesis), and (3) layered mantle convection with a compositional barrier to convection in the lower mantle. The results given here strongly favor this layered mantle hypothesis as recently set forth by Kellogget al. <1999> and van der Hilst and Karason <1999>. ¿ 2001 American Geophysical Union