Artificial neural network with radial basis function (RBF) was explored for prediction of the vapor liquid equilibrium (VLE) data. Four typical binary systems, representing different deviations from ideality (in mixing) were considered. For each type of binary system, a set of experimental VLE data from Dechema series was used to train the network. The rest of the available experimental data was used for testing the predictive capability of the network. VLE data for two ternary systems were predicted utilizing a similar network. Three sets of binary data include all the binary combinations possible with three components of the ternary system. These three sets were used to train the network for each ternary system. UNIQUAC model stores the VLE data through regression of the model parameters. The same set of VLE data can be reproduced utilizing thermodynamic models. The accuracy of such reproduction (as reported in Dechema series) is compared with the accuracy of prediction of the same set of data by RBF network. The comparison is presented here for all the four binary systems and the ternary system. The computational (storage) requirements for the networks were also evaluated. (C) 2003 Published by Elsevier Science Ltd.