Abstract: Due to the problem that the efficiency and accuracy of original spectral clustering is unstable because of its initialization sensitive, this paper presents an improved spectral clustering algorithm.The improved algorithm first constructs Laplacian matrix and its spectral eigenspace, then introduces the particle swarm optimized (PSO) FCM algorithm to find the optimum swarm and use the swarm as initial cluster centers to solve the sensitive problem.Experiment shows that the clustering result of this improved algorithm is more stable, the efficiency and accuracy of this algorithm in high-dimensional data sets are better than original spectral clustering algorithm.