Ensemble Clustering(EC) is one of the key means to solve data mining problems, but the existing EC methods rarely consider the various noises that may damage the clustering structure and reduce the clustering performance. To solve this problem, an Improved Spectral Ensemble Clustering(ISEC) method is proposed. Firstly, the clustering problem is modeled as a graph partitioning problem of coincidence matrices derived from inputting multiple Basic Partitions(BPs). Then, The ISEC method learns to obtain the low rank representation of the covariance matrix, and carries on the spectral clustering to improve the clustering performance. Finally, the optimization solution is carried out by the enhanced Lagrange multiplier method, so as to obtain the final clustering result. The simulation results on several real data sets show that the clustering performance of ISEC method is better than that of most existing clustering methods.