The distribution of ferric iron among the phases of upper-mantle rocks, as a function of pressure (P), temperature (T) and bulk composition, has been studied using ⁵⁷ Fe Mössbauer spectroscopy to determine the Fe ³⁺ /ΣFe ratios of mineral separates from 35 peridotite and pyroxenite samples. The whole-rock Fe ³⁺ complement of a peridotite is typically shared approximately evenly among the major anhydrous phases (spinel and/or garnet, orthopyroxene and clinopyroxene), with the important exception of olivine, which contains negligible Fe ³⁺ . Whole-rock Fe ³⁺ contents are independent of the T and P of equilibration of the rock, but show a well-defined simple inverse correlation with the degree of depletion in a basaltic component. Fe ³⁺ in spinel and in both pyroxenes from the spinel Iherzolite facies shows a positive correlation with temperature, presumably owing to the decrease in the modal abundance of spinel. In garnet peridotites, the Fe ³⁺ in garnet increases markedly with increasing T and P, whereas that in clinopyroxene remains approximately constant. The complex nature of the partitioning of Fe ³⁺ between mantle phases results in complicated patterns of the activities of the Fe ³⁺ -bearing components, and thus in calculated equilibrium f _O2 , which show little correlation with whole-rock Fe ³⁺ or degree of depletion. Whether Fe ³⁺ is taken into account or ignored in calculating mineral formulae for geothermobarometry can have major effects on the resulting calculated T and P. For Fe-Mg exchange geothermometers, large errors must occur when applied to samples more oxidized or reduced than the experimental calibrations, whose f _O2 conditions are largely unknown. Two-pyroxene thermometry is more immune to this problem, and probably provides the most reliable P—T estimates. Accordingly, the convergence of P—T values derived for a given garnet peridotite assemblage may not necessarily be indicative of mineral equilibrium. The prospects for the calculation of accurate Fe ³⁺ contents from electron microprobe analyses by assuming stoichiometry are good for spinel, uncertain for garnet, and distinctly poor for pyroxenes.