Abstract: Considering the characteristics of medical image, a novel medical image retrieval algorithm based on texture is proposed. Combining with the multi-scale and multi-directional characteristics of non-separable and directional filter banks, the contourlet transform can effectively capture the contours of an image and depict them sparsely. The algorithm proposed takes fully the advantage of statistical attribution of image's contourlet transform coefficients to form the eigenvector for depicting textural feature and retrieval. Retrieval tests were carried out for a CT image database with the algorithm, and the results were compared with that by the retrieval algorithm based on Harr wavelet transform. It was found that the precision of retrieval in contourlet transform domain is higher than that in wavelet transform domain, thus verifying the higher retrievability of the algorithm.