Henry's law constant is an important property for predicting the solubility and vapor-liquid equilibrium. Usually, Henry's law constants increase as temperature and salt concentration increase and polynomial correlations are commonly used to model these effects. In this article, the artificial neural network (ANN) method is used for modeling the Henry's law constant dependence on temperature and salt concentration, with methyl ketones in aqueous sodium sulfate solutions chosen for the study. In the first part, one network is used for each system. The network topology is optimized and the 2-2-1 architecture is found to be the best. The network satisfactorily estimates the Henry's law constants of all systems in the study with an average relative deviation (ARD) of less than 1% for all systems, which is comparable to available correlations. In second part, which is based on the results of the first part, an ANN is designed for all systems. The new network has a 3-2-1 topology, giving an ARD of correlation of less than 1% and ARD of prediction, depending on systems and data availability, of less than 3.5%. The predictive ability is the most important advantage of the 3-2-1 ANN compared to 2-2-1 ANN and other correlations. (C) 2008 Elsevier B.V. All rights reserved.