In order to improve the accuracy of apple grading in digital image processing system, the multi-feature information was extracted for describing the apple features. However, this method may result in information redundancy and so on. So, the principal component analysis (PCA) was used to carry out information fusion of the feature parameters, and with the aid of Wilks Λ statistic the principal components (PC) which could promote grading results were selected. Then some features used in grading were selected based on the contribution rate to selected PC. The results of Fisher discriminate analysis (FDA) showed that the grading effect corresponding to the selected features was better than that of all features, and the grading accuracy and the cross-validation accuracy rose by 2.0% and 1.5%, respectively.