Abstract: In this paper a new kind of automated fault diagnosis knowledge acquistion algorithm is proposed based on modified ant colony algorithm. The problem of fault identification and classification is translated to a constrained optimized clustering problem under certain conditions in this algorithm. And a modified ant colony algorithm, based on multi-agent cooperation and learning, is applied to solve this clustering problem. It is used to the process of fault identification and classification for fault diagnosis of a chemical reactor. The results show that the algorithm has the advantages of high parallel, high effective of computing, rapid convergence, robust and credibility of the identification result.