The purpose of this paper is the formulation and implementation of a neural network for the evaluation of the thermodynamic models UNIQUAC and UNIFAC, so that their performance in the calculation of vapor-liquid equilibria, a very important aspect of chemical process design, can be estimated. A multi-layer network with feedforward connections is used. Each processing unit is a semi-linear neuron (the activation rule is a sigmoid function) and synapses do not exist between elements of the same layer. The training and prediction examples are obtained from vapor-liquid equilibrium data for several isothermal binary systems composed of hydrocarbons and alcohols. The temperature of the system, UNIFAC groups and acentric factor of the compounds and mean pressure deviation of the UNIQUAC and UNIFAC models are used in the examples. For the implementation of the network, the software NeuralWorks Professional II/Plus, NEURALWARE, Inc. is used. After training, satisfactory agreement was found between the answers calculated by the network and the output patterns presented to it. The success of the implementation is demonstrated by testing its predictive capability.