In the last 50 years, several studies have been performed on the measurement and prediction of hydrate forming conditions for various gas mixtures and inhibitors. Yet, the correlations presented in the Literature are not accurate enough and consider most of the time, simple pure gases only and their mixtures. In addition, some of these correlations are presented mainly in graphical form, thus making it difficult to use them within general computer packages for simulation and design. The purpose of this paper is to present a comprehensive neural network model for predicting hydrate formation conditions for various pure gases, gas mixtures, and different inhibitors. The model was trained using 2387 input-output patterns collected from different reliable sources. The predictions are compared to existing correlations and also to real experimental data. The neural network model enables the user to accurately predict hydrate formation conditions for a given gas mixture, without having to do costly experimental measurements. The relative importance of the temperature and the different components in the mixture has also been investigated. Finally, the use of the new model in an integrated control dosing system for preventing hydrate formation is discussed. (C) 1998 Elsevier Science B.V. All rights reserved.