The goal of this study is to analyse how the thermal diffusion process is dependent on molecular parameters when describing a fluid mixture. To estimate the associated transport coefficient, which is the thermal diffusion factor alpha(T), a non-equilibrium molecular dynamics algorithm has been applied on equimolar binary mixtures of Lennard-Jones (U) particles in supercritical conditions. Firstly, it is shown that this model is able to correctly estimate alpha(T) for simple alkane mixtures, provided there are a sufficient number of particles and long enough simulations. Then, using various mixing rules, the separate influences of the mass, the moment of inertia, the atomic diameter and the interaction strength have been studied. Results indicate that the molar fraction of the component, having the smallest mass and moment of inertia as well as the biggest radius and the strongest potential, tends to increase in the hot area. Elsewhere, simulations for various cross-interaction parameters show that alpha(T) is extremely sensitive to the intermolecular pair potential between unlike particles. Finally, results on methane/normal alkane mixtures indicate that a simple sum between the separate contributions provides a reliable evaluation Of alpha(T) only when the molecular parameter ratios between the two components are close to 1. (C) 2003 Elsevier Science B.V. All rights reserved.