Abstract: To eliminate the gap between the high-level semantics and the low-level features of images, a new method based on co-occurrence matrix is proposed to extract image texture feature. Combining the image's statistical features in the frequency domain with its spatial distribution attributes, the method extracts the local frequency information on image by wavelet transforming. Then the image's global structural characteristics are integrated with its wavelet properties to construct the wavelet-gray co-occurrence matrix so as to extract the image's texture features for the retrieval of medical images. The results of comparative test showed that the wavelet-gray co-occurrence matrix is superior in medical image retrieval in comparison to other ways where the co-occurrence matrix is separated from wavelet features in applications.