Gustafson- Kessel (GK) clustering algorithm can be used to search for hyper ellipsoid,plane and lineardata effectively,but it is still sensitive to initial clustering center and easy to fall into a local optimum.In order tosolve this problem,this paper presents an improved GK clustering algorithm,which adopts the artificial fish- swarm(AFS) algorithm to initialize clustering centers according to the similarity between clustering and searching food offish school.Then,this algorithm is employed to perform a simulation based on artificial data and IRIS data.Theresults show that the improved algorithm can find the clustering structure of datasets effectively,and it possesses abetter classification performance than GK clustering algorithm.