Abstract: A spectral bisection algorithm based on multidimensional eigenvector is proposed for the division of web communities in large-scale complex networks, according to different community structures. It identifies those communities by way of the multidimensional eigenvectors of network's connection matrices, and then the effects of key parameters on the division results are analyzed through simulation so as to determine the parametrical values which make sure that the results of division are optimum. And the number of communities being identified can thus be confirmed by the factors integrating the threshold values of multidimensional eigenvector with the total number of communities in a network. The results of simulation done in a locally typical WWW evolution model demonstrated that the algorithm proposed enables the efficient division of many communities in relevant network where the clustering characteristics aren't so obvious. In addition, the algorithm is adaptable to the networks with various clustering characteristics, thus showing actually its high applicability to networks.