Modeling and prediction of protein solubility is a key to developing the protein crystal growth and crystallization process. In this paper a back propagation network was used for predicting the solubility of protein in lysozyme-NaCl-H2O system. It was found that properly selected and trained neural network could fairly represent the dependence of protein solubility on the pH, salt concentration, and temperature. The RMSD (root mean square deviation) for prediction of the solubility of lysozyme in lysozyme-NaCl-H2O system was 0.07% by the artificial neural network (ANN) method, which is better than that of with thermodynamic models. The ANNs have been proven to be an effective tool for correlation and prediction of protein solubility in protein-salt-water system. (C) 2003 Elsevier B.V. All rights reserved.