The neural network was applied to the estimation of catalyst deactivation by taking, as an example, methanol conversion into hydrocarbons over ion-exchanged dealuminated mordenites. In the first series, it was attempted to estimate the deactivation rate constant, k(d) defined in -dA/dt = k(d)A where A is the degree of conversion, from the amount of strong acid sites and the catalyst composition such as the Si/Al ratio and the degree of ion exchange. The estimated rate constant agreed well in most cases with the experimentally obtained constant. The most serious exception was Ba ion-exchanged dealuminated mordenite which experimentally exhibited the slowest deactivation. Better agreement was obtained when the first-order reaction rate constant was used as A in the above equation instead of the degree of conversion. In the second series, it was shown that the neural network has a strong ability to extrapolate the catalyst decay curve even without knowing catalyst composition and properties, especially when the first-order reaction rate constant was used to represent the catalyst activity. All of these results clearly demonstrate that the neural network is a powerful tool to estimate the deactivation behaviour of catalysts. (C) 2004 Elsevier B.V. All rights reserved.