Adsorption and desorption isotherms for corn starch powders were determined at 30, 45 and 60 degrees C using a gravimetric technique. Samples were equilibrated in desiccators containing sulphuric acid solutions of known water activity (0.05-0.95), and placed in temperature-controlled cabinets for approximately 10 days. The data obtained were fitted to several Halsey, Oswin, Henderson, Modified-BET and Smith GAB, Ferro-Fontan, Peleg and a new sorption model of corn starch powder on BP neural network were established. Analysis showed within the temperature range investigated, GAB, Peleg and Henderson models better describes the experimental data for corn starch powder. BP neural network model not only accommodated temperature and water activity parameter, but also is better than other mathematical models. (c) 2006 Elsevier Ltd. All rights reserved.