Following a detailed analysis of the experimental data on thermodiffusion coefficients of numerous binary hydrocarbon mixtures from over 90 experiments, four new algebraic expressions are proposed. As per these formulations, the thermodiffusion coefficient of binary n-alkane mixtures can be expressed as a function of the chemical composition of the mixture and the mixture properties such as density and viscosity. Detailed experimental validation is presented using four n-alkane series containing a wide range of combinations of n-alkanes. Additionally, comparison with recently proposed neural network model and a model based on the principles of non-equilibrium thermodynamics is also presented. It has been found that the proposed algebraic models are simple in formulations, are evaluated with least computational effort and yet have a very high accuracy in predicting the thermodiffusion coefficients. (c) 2012 Canadian Society for Chemical Engineering