To match reliably point pairs,a matching algorithm based on the probabilistic relaxation of the spectral correlation is proposed. Firstly,the initial matching probabilities are obtained from the shape context of the two point sets. Then,two proximity matrices are defined from the point sets respectively,and the spectral correlation of the matrices as the initial support is acquired by the singular value decomposion (SVD). Finally,the matching of the two point sets is implemented by using the method of probabilistic relaxation. Experimental results show the high accuracy of the algorithm.