This paper deals with the proposal of a predictive method for Margules parameters using reconstruction-learning neural network (NN). The input layers in the NN method are critical volume, acentric factor, dipole moment, entropy of vaporization and electronegativity of components I and 2. The number of Margules parameters used for evaluating the weight matrix in the NN method is 872, and the obtained correlation coefficient is R-2=0.8537. The Margules parameters not used as learning data were predicted for 17 binary systems. The vapor-liquid equilibria were then predicted for 17 binary systems using these Margules parameters in combination with Riedel vapor pressure constants predicted by the NN method proposed previously. We observed a high degree of similarity between experimental and predicted vapor compositions. (c) 2007 Elsevier B.V. All rights reserved.