The price of beef is determined by the carcass grade,and beef grade usually evaluated subjectively based on marbling level. It is a good method to determine marbling level with computer vision instead of person. The method assessing beef grade based on image processing was studied in this paper.The image of longissimus dorsi muscle was segmented using threshold, dilation and erosion, and the parameters of marbling that extracted from dorsi muscle were described using moment. The model predicting grade of marbling based on support vector machine was obtained. The experiment indicated that the predicting accuracy was 95.9%, 89.2%,93.2%, 100% as grade 1, 2, 3, 4 respectively.