In order to establish predicting models for beef marbling grade, one hundred and thirty-five beef rib-eye samples were collected from a beef processing factory. A three-grader panel was employed to assign each sample a marbling score according to the official standard cards. Then images of each sample were captured with a CCD camera. The beef marbling images were separated and the indicators such as ratio of fat area, the total number of fat particles and numbers of the big and the small fat particles were obtained by using image processing technologies. After fractal dimensions of marbling images was calculated through the varying megascopic degree method, a multiple linear regression model and a multiple polynomial model for the prediction of marbling score were derived based on the obtained fractal dimensions and image feature parameters. Validation results showed that the correct percentage of the marbling grade predicted by the multiple linear model and by the multiple polynomial model were 75% and 87.5%, respectively, indicating that beef marbling grades can be determined by using fractal dimension and image processing method.