Clustering algorithms have been useful for fault detection of process systems. Among many algorithms, ART2 neural networks have provided good results for fault detection and identification. In this article, a new algorithm for fault detection is proposed that is based on clustered classes from ART2 neural networks. An index for detecting changes of process state is defined, to represent differences between recent class distribution and historical class distribution. The proposed algorithm was applied to detection of state change of an industrial wastewater treatment process and was shown to be effective in detecting a change in the process.