A mutual-diffusion-coefficient model based on local-composition concept is developed. An expression for the calculation of concentration-dependent intra-diffusion coefficients is also established by considering the formation of clusters. With the import of diffusion coefficients at infinite dilution, self-diffusion coefficients and viscosities of pure components and mixtures, mutual-diffusion-coefficients of binary fluid mixtures can be predicted. The total average relative deviation of predicted values with respect to experimental data is 6.0% for 45 binary systems including those containing associative component. Results indicate that this model is superior to some currently used methods.