One major difficulty in the shape-based image retrieval area is how to keep good results after translation, rotation and scaling of the image. A novel method based on Krawtchouk moment invariants is proposed for shape-based image retrieval. Firstly, the image was transformed into gray image. Then, the low-order moments of the image were extracted. Finally, 16 low-order moment invariants were selected as image feature vectors. The similarity distance measure between feature vectors adopted Euclidean distance function. According to the similarity degree, similar images were retrieved and image retrieval was completed once. Experimental results were given. A comparison was made between the proposed method and the image retrieval method using geometric moment invariants or Zernike moment invariants. The experimental results show that the proposed method is more effective in image retrieval and improves its accuracy. Compared with above-mentioned two methods, the recall increases by 21.52% and 7.6%, respectively, while the precision increases by 16.25% and 6.25%, respectively. So the proposed method is more suitable for shape-based image retrieval.