Abstract: A fabric retrieval algorithm using Scale-Invariant Feature Transform (SIFT) and Vector of Locally Aggregated Descriptors (VLAD) feature encoding is proposed in this paper. Firstly, SIFT features of images are extracted to represent images. However, different images usually contain different numbers of SIFT feature points. That causes a problem that feature dimensions of two different images are inconsistent so that the similarity between the images cannot be directly calculated. To solve this problem, the VLAD feature encoding is further implemented to ensure the consistency of feature dimensions of different images, while the feature representation ability of SIFT feature is also improved. The VLAD encoding includes two steps. First, learning a visual dictionary by using the K-means clustering algorithm. Second, local aggregation of eigenvectors. The local aggregation step contains three sub-steps: 1) calculating the residuals between SIFT feature vectors and corresponding visual words in the image ; 2) summarizing up the residuals corresponding to each visual word; 3) concatenating the residual sum values of each visual word were to obtain the VLAD code of the image. In this paper, the 10-time average of Cumulative Match Characteristic (CMC) curve is used as the performance measurement. The experiment results show that the proposed method is able to improve recognition speed and acquire a high identification rate, i.e., the average rank-1 identification rate is 95.03 %.