Abstract: In order to improve both speed and accuracy of image database retrieval, particularly for large image database, a two-level query technique based on HSV colorspace is proposed in this paper. Firstly, we narrow the scope of inquery by the use of color moments which is simple and effective. Then, we split images into 4×4 overlapping sub-blocks and extract color histograms of sub-blocks for the second-level retrieval of image database to obtain a more accurate searching results. The benefits of such a block are not only obtaine the spatial distribution of image information, but also highlight the main part of the image and fully restrict the scope of the background. This technique reduces the computational complexity compared to an independent sub-block. As shown in the experiments, with an example image given by the user, good retrieval results can be obtained by two-level query technique with color features.