According to the uncertainty and low accuracy of apple grading relying on the single feature, key features such as size, shape, color and defect which could show the apples appearance quality were extracted and a decision fusion method based on D—S evidence theory was proposed. Firstly, the apples were graded according to each of the four features utilizing human experience and neural network. Then, the former grading results were used as evidences to achieve the decision fusion in order to increase the grading reliability and accuracy. Finally, 80 apple samples were used to test grading effect. The experimental results showed that five apples were misjudged and the grading accuracy reached to 92.5%. The proposed method had good performance on classification rate and stability compared to the grading method based on single feature. 