An apple grading method based on decision fusion of discriminant tree and improved support vector machine was proposed. The method of discriminant tree classification was used to classify fruit diameter, defect area and color, and the particle swarm optimization (PSO) was used to optimize the SVM classification model. The high dimensional features, such as fruit shape, texture and maturity, were used to classify, and the kernel principal component analysis (KPCA) was used to reduce the dimension. While, the concept of decision fusion was introduced to comprehensively evaluate the sample level combined with single feature. The results showed that the method was feasible, and its classification accuracy was more than 98%, which can be used for apple grading effectively.