Abstract: Determining the number of sources is very important to high resolution algorithm.To solve this problem, the general way is to count the number of eigenvalues satisfying the threshold calculated from the sample covariance matrix by some criteria.In this paper, the character of eigenspace is concerned in both eigenvalues and eigenvectors for the sources spaced closely.As a result, a new approach is proposed, which determines the number of sources by synthetic use of eigenvectors as well as eigenvalues.Performance of the new approach is superior to eigenvalue-based approaches for the sources spaced closely(less than half power beam width), and equal to MDL(Minimum Description Length) otherwise.Simulation results are presented to confirm the conclusion.