We review the experimental and theoretical constraints on the distribution of trace elements in the transition zone and lower mantle, with particular emphasis on refractory lithophile elements and H₂O. We begin with a summary of element partitioning between high-pressure minerals of transition zone and lower mantle and coexisting silicate melts. Mineral–melt partitioning of trace elements in the deep Earth obeys the elastic strain model with partition coefficients showing a near-parabolic dependence on ionic radius for cations of fixed charge. Experiments also indicate that many elements (e.g. Zr, Hf, U, Th) which are incompatible in upper mantle minerals are compatible in Mg- and Ca-perovskites of the lower mantle. The high partition coefficients of, for example, 3+ and 4+ cations into the Ca-site of Ca-perovskite arise from the ease with which excess charge is compensated in the perovskite structure by creation of cation vacancies. One important implication is that, despite being a volumetrically minor phase, Ca-perovskite contains most of the heat-producing elements Th and U in the deep Earth, as well as being the principal host of the rare earth elements. Measured mineral-melt partition coefficients for deep mantle minerals also indicate that there cannot be large volumes of a majoritic or perovskitic reservoir isolated in the lower mantle since the ‘magma ocean’ stage of early Earth history. Given this constraint, any such region would have geochemical characteristics similar to those of the HIMU component of oceanic basalts while downward migration of dense melts in equilibrium with Ca-perovskite could, in principle, lead to formation of a complementary reservoir, unradiogenic in Pb and with a subchondritic ¹⁴²Nd/¹⁴⁴Nd ratio. Water-solubility measurements on deep mantle minerals demonstrate that wadsleyite and ringwoodite of the transition zone can dissolve much larger amounts of water (>2 wt.%) than the low-pressure olivine polymorph (~5000 ppm). Maximum solubility does not, however, prove that high H₂O contents are present in the Earth and the seismologically determined sharpness of the 410 km discontinuity constrains the H₂O content at this depth to about 400 ppm by weight. Water-solubility measurements on the principal materials of the lower mantle are currently very scattered and there is a need for improvement of both synthesis and analytical methods. It is clear, however, that water is more soluble in the transition zone than in the lower mantle. These observations and mineral-melt partition coefficients for trace elements are not inconsistent with the transition zone ‘water filter’ hypothesis by which depleted upper mantle is generated by hydrous partial melting at 410 km depth. The water-storage capacity of the upper mantle and the H₂O contents of mid-ocean ridge basalts are, however, difficult to reconcile with this hypothesis.