We propose a new method of feature extraction in orderto improve the effective of image retrieving by using apartial Joint Photographic Experts Group (JPEG)compressed images algorithm. Prior to that, we prune theimages database by pre-query step based on coloursimilarity, in order to eliminate image candidates. Ourfeature extraction can be carried out directly to JPEGcompressed images. We extract two features of DCTcoefficients, DC feature and AC feature, from a JPEGcompressed image. Then we compute the Euclideandistances between the query image and the images in adatabase in terms of these two features. The image querysystem will give each retrieved image a rank to define itssimilarity to the query image. Moreover, instead of fullydecompressing JPEG images, our system only needs to dopartial entropy decoding. Therefore, our proposed schemecan accelerate the effectiveness of retrieving images.According to our experimental results, our system is notonly highly effective but is also capable of performingsatisfactoril.