Abstract: In view of problems that coal-rock interface sensor has complicated structure, low reliability and universality in existing coal-rock identification method, the paper proposed a solution of coal-rock interface identification system based on machine vision. It gave general structure of the system, analyzed its working principle and focused on discussion of image feature selection and classifier design. The system extracts 22 textures of coal-rock images according to the theory of gray-level co-occurrence matrix, employs method of increasing l and decreasing r to search the optimized features, and builds coal-rock classifier model by linear function recognition. The experiment results show that the coal-rock classifier model of the system has stable performances and strong identification ability.