The pressure dependence of the thermal expansion coefficient, &agr;, previously reported as (∂ln&agr;/∂lnV)T=5.5¿0.5 by Chopelas and Boehler [1989> is refined, using systematics in the volume dependence of (∂T/∂P)s measured for a large number of materials at high pressures and high temperatures. Since (∂ln(∂T.∂P)s/∂(V/V0))T is found to be constant and material independent over a very large compression range, (∂ln&agr;/∂lnV)T is proportional to the compression, V/V0. We find &agr; decreases by a factor of 5 for MgO throughout the mantle, reaching a value of 1.0⋅10-5 K-1 at its bottom. Densities of perovskite (PV) and magnesiow¿stite (MW) are calculated for lower mantle conditions using our new &agr;(P,T), a room temperature finite strain equation, and recent data on the Mg-Fe partitioning in the PV-MW system. Both minerals have nearly identical densities to those of PREM throughout the entire lower mantle, which allows variable PV:MW ratios. A lower mantle made entirely of PV with a molar ratio of Mg:Fe of 88:12 would be about 0.11 g/cm3 or 2.5% denser than this mixture, but this density would just be within the uncertainty in PREM. A change in chemistry at 660 km depth to a PV mantle requires a thermal boundary which would improve the match in the densities between PV and PREM. These density agreements therefore preclude evaluation of a mineralogical model for the lower mantle using density comparisons. Recent measurements on melting of Fe, FeO, and FeS, however, suggest temperatures at the core-mantle boundary below 3500 K, which tends to favor a geotherm without a large thermal boundary at 660 km depth. ¿ American Geophysical Union 1992