Abstract: In computer vision, the key of 3 D reconstructing is to figure out the depth information of a scene from its images. A new method for depth from defocus images is proposed. Two images of the same scene are taken with different defocus by changing the distance between lens and image plane using telecenter lens. The observed two defocus gray level images are converted into gradient image. The ratio P e between the edge region and the entire gradient image is calculated using moment preserving principle. According to P e of two images, the depth of scenery is determined. Experiments results show that this method is effective. The error of this method is analyzed.