Thermal diffusion is important for the study of composition variations in hydrocarbon reservoirs, and it can either enhance or weaken the separation in mixtures. We present a new model for the prediction of thermal diffusion coefficients in binary mixtures of reservoir fluids using the thermodynamics of irreversible processes. The model needs equilibrium properties of mixtures and energy of viscous flow. Equilibrium properties are obtained from the volume translated Peng-Robinson equation of state, and the energy of viscous flow is estimated from viscosity. The model has been applied to predict thermal diffusion coefficients of several mixtures consisting of nonhydrocarbon and hydrocarbon fluids. Comparisons of theoretical results with experimental data show a good performance of the model except in the near-critical region where all existing models are deficient. In particular, the predicted sign of thermal diffusion coefficients is consistent with experimental observations in systems investigated here, namely, C-1/C-3, C-1/C-4, C-7/C-12, C-7/C-16, Ar/CO2, N-2/CO2, H-2/N-2, H-2/CO2, and C-1/CO2, except in C-1/N-2, in which the values of thermal diffusion coefficients are extremely low, We have also modified some of the earlier models, such as the Kempers, Haase, and Rutherford models, which are based on phenomenological and kinetic approaches. In general, our model has been found;to be most reliable and represents a significant improvement over the earlier models.