The paper presents a method of image recognition for apple picking robot at night. In the stage of image acquisition, comparing different lighting effects of light source, the proposed method selected the incandescent lamp as artificial lighting. And two lamps were placed in different sites to weaken shadow with 60° angle. In the stage of image segmentation, comparing the image recognition methods adopted in daytime, the improved R-G image recognition method was proposed. Referring to reflective areas in the surface of apples, twice segmentation method was proposed to separate reflective areas from images to repair segmented images above. The first segmentation aimed to determine the areas in where apples were located. The second segmentation aimed to determine the reflective areas in the determined areas of first segmentation. Taking no account of sheltered and adhesive apples, the rate of identification on the basis of the statistic of 60 images acquired at night can reach to 83.7%. This paper verified the ability of image recognition for apple picking robot at night and discussed the feasibility of apple picking at night.