Abstract: In this paper a self-made automatic coal gangue sorting device was employed to investigate the automatic image recognition technology on coal and gangue. The gray features of coal and gangue images and their texture features based on gradation co-occurrence matrix were introduced. Besides, the normalized eigenvectors are constructed by using the energy, entropy, contrast, and correlation of the grayscale features and texture features. Finally the BP neural network was used to identify and clarify, and the influence of different learning rates on the recognition rate was experimentally analyzed. The results showed that the comprehensive classification method of texture and gray features based on BP neural network improved the recognition rate of coal and gangue; suitable learning rate improved the learning speed of BP neural network and the recognition rate which reached to 87.5%.