As a class of porous materials, metal-organic frameworks (MOFs) show promise for the adsorption-based separation of mixtures of gases. The design of any process involving selective adsorption requires knowledge of mixture adsorption isotherms. Ideal adsorbed solution theory (IAST) predicts mixture adsorption equilibria using only single-component data, thereby minimizing the need for experimental adsorption data. In this work we perform a systematic study of the applicability of IAST to MOFs by using grand canonical Monte Carlo (GCMC) simulations to investigate the suitability of IAST for the prediction of the adsorption of mixtures of molecules of differing sizes, asphericities, and polarities in a range of structurally different MOFs. We show that IAST is generally accurate for MOFs. Where we find IAST is less accurate, deviations result from both mixture effects, in the form of nonidealities in the adsorbed phase, and characteristics of the adsorbent structures. In terms of the MOF structure, departures from IAST are a consequence of heterogeneities both on the scale of the unit cell and on shorter length scales, whereby competition for adsorption sites has a strong influence.