An artificial neural-network analysis was applied to the prediction of conversions and yields in catalytic reactions : decomposition of NO into N-2 and O-2 over Cu/ZSM-5. With a proper network structure and training sets, the evolved neural-network system provides plausible activities of catalysts from a discrete and finite number of experimental data. A two- or three-dimensional graphical expression of the calculated data seems to be useful to reveal the relationship between the catalytic activity and reaction conditions. Therefore, they can be utilized in finding optimum reaction conditions or in examining effects of various factors on catalytic reaction.