Multicomponent diffusion is ubiquitous in (bio)chemical processes. The Maxwell-Stefan (M-S) framework provides a sound theoretical basis for describing transport diffusion as it correctly accounts for the gradient in chemical potential as driving force. Unfortunately. M-S diffusivities D-ij cannot be measured directly in experiments. The use of predictive models based on easily measurable quantities like Fick-or self-diffusivities in diluted systems is therefore desirable. In this study, equilibrium molecular dynamics (EMD) simulations are used to study M-S diffusivities in liquid mixtures containing n-hexane, cyclohexane and/or toluene. Predictive models for estimating M-S diffusivities in ternary systems are investigated. The following analysis are carried out. First, these predictive models are used to calculate the self-diffusivity in the infinite dilution limit using the well-known Vignes approximation. The predicted self-diffusivity is compared to the self-diffusivity directly calculated from EMD simulations. Second, we investigated the quality of the Vignes approximation using diffusivities obtained from EMD simulations. Third, we directly compared the predictive models for D-ij(xk-1) with EMD simulations. Our results show that: (1) predicted self-diffusivities are not very sensitive to the choice of the predictive model for D-ij(xk-1); (2) the Vignes equation results in only reasonable predictions for M-S diffusivities, yielding errors of113% on average; (3) the interaction between solutes and solvent cannot be neglected in predictive models for D-ij(xk-1); (4) present predictive models for calculating D-ij(xk-1) from binary data results in errors of 8% for the systems under investigation. (C) 2010 Elsevier B.V. All rights reserved.