China's apple export is far lower than that of Europe, America and some other countries. The reason is that the post harvest classification technology of apple in China is underdeveloped. Since the traditional apple classification technology does not solve the ordered information of classification targets in classification. Therefore, in this paper, ordered partition neural network is applied as the model of apple grading. Based on the partial least square method, the features of an sample are extracted from the near-infrared spectrum of apple, meanwhile, the soluble solid content is transformed into an ordered label as the output. An ordered partition neural network model is constructed, and the apple classification based on ordered regression algorithm is studied and optimized. The results of apple grading experiment show that the prediction accuracy of ordered partition neural network can reach 88.75%.