In the research of content-based image retrieval, the former work was focused on the choice and the extraction of ocular features principally. But this kind of retrieval system cannot meet the expectations of practical use, because it can neither make full use of people?s subjectivity, nor integrate low-level image features with semantic contents of images. In order to solve this problem, this paper uses the scheme of region weight adjusted automatically and Tabu Search algorithm, presents a tabu search-based relevance feedback algorithm in image retrieval. The experimental results show that after the region weight adjusted by this algorithm. The retrieval performance is improved very obviously.