The work aims to propose a surface defect image screening method for paper-plastic composite bag based on gradient projection difference, so as to solve the problem of low efficiency and poor accuracy in defect image screening. Firstly, the image of paper-plastic composite bag was acquired and the gradient magnitude image was obtained by Sobel operator function. Then, the information of the acquired gradient magnitude image was calculated by the horizontal projection and vertical projection calculation formula, and the average value of horizontal projection and vertical projection of the image was derived. The difference in average value of gradient projection was analyzed according to the feature of abundant gradient information in the acquired defect image, and the threshold value was calculated according to the image information. The screening of the image was completed by comparing the projection difference and the threshold value. For the acquired defect image, when the threshold value of projection difference was set in the range of 0.014 6 to 0.018 9, the image of surface defect could be screened. Compared with the differential shading method, the screening correct rate of this method was improved by 10% and the image screening time was accelerated by 1.5 s. The gradient projection difference method proposed has the advantages of high correct image screening rate and fast screening speed compared with the differential shading method, which can meet the needs of image screening in industrial production.