The feature based on visual attention model is a new feature that can objectively interpret the highlevel concepts. High retrieval efficiency will be achieved and the semantic gap in image retrieval can be also reduced effectively if visual attention mechanism is adopted. An improved Itti′s visual attention model, inspired by human visual perception, is proposed. Primary component map is used to take place of intensity map, and the information of texture coarseness is brought in the improved Itti′s model. A novel image retrieval algorithm is proposed based on visual attention distribution feature. Firstly, the image is divided into 38 feature maps by the improved Itti′s visual attention model. Then, by the fixed fourbyfour grid of subregions, the visual attention distribution feature to image retrieval is constructed according to statistical distribution of the 38 feature maps which contain most of the structural information of the image. The experimental results show that the proposed method has better retrieval performance.