Abstract: The conversion rate of dehydrogenation of ethylbenzene to styrene is important for the decision-making in the management of the chemical plant. In this paper，we applied a modified radial basis function (RBF) neural networks that could endure better fault-tolerace for this problem. The modified RBF neural networks was tested by six case history data sets，i.e.，the feed of the ethylbenzene，the temperatures of the first reactor and second reactor，the output pressure of the second reactor，the ratio of steam to ethylbenzene，the selectivity of the catalyst. The experiment showed that this method could ensure high accuracy in predicting the conversion rate of the process.