Many experimental data demonstrate distinct differences between crystal-liquid partition coefficients Di measured under high temperature, anhydrous conditions and those determined at lower temperatures in the presence of H2O. In this study we develop a thermodynamic method of separating the effects of H2O from those of temperature. We then apply the method to predict partitioning of REE between clinopyroxene, garnet and silicate melt over wide ranges of temperature, pressure and H2O content. Our initial inputs are the model of Wood and Blundy (1997) for REE partitioning between clinopyroxene and anhydrous melt and the melting temperatures of diopside on the join CaMgSi2O6-H2O (Eggler and Rosenhauer, 1978). We then make the hypothesis that the effect of H2 O on the activity of REE clinopyroxene component (REEMgAlSiO6) in the melt is the same as the measured effect on CaMgSi2O6 component. This leads to predictions of REE partition coefficients for clinopyroxene coexisting with hydrous melt at any P,T and H2O content up to 45 weight %. The results agree with observed REE partition coefficients with a standard deviation which is the same as that for the anhydrous data. We conclude, therefore, that the 'H2O-effect' may, in this case, be accurately predicted. We extend the approach to garnet by using the join Mg3Al2Si3O12-H2O to estimate the effects of H2O on all 'garnet-like' components in the melt. This enables calculation of garnet-melt REE partition coefficients for melts containing up to 25% H2O. The observation that H2O influences major and trace component activities in a similar manner enables us to make some generalisations about the combined effects on partitioning of decreasing temperature and increasing water content of the melt. The relative enthalpies of fusion ΔHf of major and trace components dictate whether trace element partition coefficients increase or decrease with H2O addition: ΔHf trace > ΔHf major f gives increasing Dtrace with addition of H2O and ΔHf trace < ΔHf major leads to decreasing Dtrace with addition of H2O. Note that Dtrace is strictly the ratio of mole fractions of charge-balanced components such as REEMgAlSiO6 in solid to melt phases. In the cases considered here, however, Dtrace closely approximates D(=[trace]solid/[trace]melt) where [trace] refers to weight concentration. For clinopyroxene ΔHfREE <ΔHf major and H2O addition leads to substantial decreases in REE partition coefficients. For garnet, ΔHfREE ≈ ΔHf major and addition of H2O results in little change in DREE. The consequences are illustrated with respect to Yb contents of melts of spinel lherzolite. For partial melting under anhydrous conditions, Yb is relatively compatible in clinopyroxene and the Yb contents of the product melts remain close to 5 times source values at all F (fraction of equilibrium batch melting) between 0 and 0.25. Hydrous peridotite produces melts, because of low DYb, which are strongly enriched in Yb (up to at least 10 times source) and whose Yb concentrations depend strongly on F. This means that the low Yb concentrations observed in many island arc tholeiites cannot be due to the addition of water to fertile peridotite.