Simple empirical functions have been proposed for binary mixture properties fitting. The models are defined as rational functions of the canonical polynomials of Scheffe. The correlations are relatively easy to use and are applicable to a wide variety of properties-composition relationships. If the ternary effects are insignificant they represent the ternary data using parameters obtained from binary information with high accuracy. Additional advantages are also their good interpolating properties. In order to establish the statistical characteristics of the proposed models, non-isothermal viscosity data for eight binary mixtures were tested successfully. The mean absolute deviation between the observed and predicted values was found to be less than 1%.