Fe3+/∑Fe ratios were determined from Mössbauer spectra recorded for a series of 17 anorthite-diopside eutectic glasses containing 1 wt% 57Fe2O3 quenched from melts equilibrated over a range of oxygen fugacities from fO2~ 105 bars (Fe3+/∑Fe = 1) to 10−13 bars (Fe3+/∑Fe = 0) at 1682 K. Fe3+/Fe2+ was found to be proportional to fO2 to the power of 0.245 ± 0.004, in excellent agreement with the theoretical value of 0.25 expected from the stoichiometry of the reaction Fe2+O + 0.25 O2 = Fe3+O1.5. The uncertainty in the Fe3+/∑Fe ratios determined by Mössbauer spectroscopy was estimated as ± 0.01 (1σ) from the fit of the data to the theoretical expression, which is significantly less than that quoted for previous measurements on silicate glasses; this results from fitting the spectra of a large number of systematically varying samples, which allows many of the ambiguities associated with the fitting procedure to be minimized. Fe3+/∑Fe ratios were then determined for samples of the anorthite-diopside eutectic composition equilibrated at selected values of fO2, to which up to 30 wt% Fe2O3 had been added. Fe3+/∑Fe was found to vary with ∑Fe (or FeOT), but both the 1 wt% and high FeOT data could be satisfactorily fit assuming the ideal stoichiometry (i.e., Fe3+/Fe2+ ∞fO2 1/4) by the inclusion of a Margules term describing Fe2+-Fe3+ interactions. The large negative value of this term indicates a tendency toward the formation of Fe2+-Fe3+ complexes in the melt. The resulting expression, using the ideal exponent of 0.25, gave a fit to 289 Fe3+/∑Fe values, compiled from various literature sources, of similar quality as previous empirical models which found an exponent of ~0.20. Although the empirical models reproduce Fe3+/∑Fe values of glasses with high FeOT reasonably well, they describe the data for 1 wt% FeOT poorly. The non-ideal values of the exponent describing the dependence of Fe3+/∑Fe on fO2 at high FeOT are an artifact of models that did not include a term explicitly to describe the Fe2+-Fe3+ interactions. An alternative model in which Fe in the silicate melt is described in terms of three species, Fe2+O, Fe3+O1.5, and the non-integral valence species Fe2.6+O1.3, was also tested with promising results. However, at present there is no model that fits the data within the assessed accuracy of the experimental measurements.