Abstract: A two-step progressive texture retrieval algorithm for multi-source remotely sensed imagery based on Contourlet transform and spectral histogram is proposed.Firstly,the Contourlet transform is applied to decompose texture features of remotely sensed imagery at different scales and different directions.Then,low-pass signal and high-pass signals are adopted to realize elementary retrieval and exact retrieval relatively.The experiments on texture database obtained from USC and QucikBird images show that the proposed algorithm not only utilizes the advantages of Contourlet on multi-scale and multi-direction texture feature extraction,but also makes full use of the efficiency of spectral histogram on distributed statistical feature representation.On the aspect of precision and efficiency,Contourlet transform outperforms wavelet transform.The proposed method provides a powerful tool for texture retrieval of remote sensing image.