The partitioning of Fe and Mg between coexisting garnet and olivine has been studied at 30 kb pressure and temperatures of 900 ° to 1,400 °C. The results of both synthesis and reversal experiments demonstrate that KD (= (Fe/Mg)gt/(Fe/Mg)OI) is strongly dependent on Fe/Mg ratio and on the calcium content of the garnet. For example, at 1,000 °C/30 kb, KD varies from about 1.2 in very iron-rich compositions to 1.9 at the magnesium end of the series. Increasing the mole fraction of calcium in the garnet from 0 to 0.3 at 1,000 ° C increases KD in magnesian compositions from 1.9 to about 2.5. The observed temperature and composition dependence of KD has been formulated into an equation suitable for geothermometry by considering the solid solution properties of the olivine and garnet phases. It was found that, within experimental error, the simplest kind of nonideal solution model (Regular Solution) fits the experimental data adequately. The use of more complex models did not markedly improve the fit to the data, so the model with the least number of variables was adopted. Multiple linear regression of the experimental data (72 points) yielded, for the exchange reaction: 3Fe2SiO4+2Mg3Al2Si3O12 olivine garnet ⇌ 2Fe2Al2Si3O12+3Mg2SiO4 garnet olivine ΔH ° (30kb) of -10,750 cal and ΔS ° of -4.26 cal deg-1 mol-1. Absolute magnitudes of interaction parameters (Wij) derived from the regression are subject to considerable uncertainty. The partition coefficient is, however, strongly dependent on the following differences between solution parameters and these differences are fairly well constrained: WFeMg ol-WFeMg gt≃ 800 cal WCaMg gt-WCaFe gt≃ 2,670 cal. The geothermometer is most sensitive in the temperature and composition regions where KDis substantially greater than 1. Thus, for example, peridotitic compositions at temperatures less than about 1,300 ° C should yield calculated temperatures within 60 °C of the true value. Iron rich compositions (at any temperature) and magnesian compositions at temperatures well above 1,300 °C could not be expected to yield accurate calculated temperatures. For a fixed KDthe influence of pressure is to raise the calculated temperature by between 3 and 6 °C per kbar.