Abstract: In order to improve accuracy of information of mineral regions from the images generated in intelligent X-ray refuse identification process, the X-ray images of graphite, quartz, kaolinite and mortmorillonite, four kinds of main minerals in coal and refuse, are respectively obtained first, and then the high-energy and low-energy X-ray images of coal and refuse are respectively produced.The images after being converted to grey-level ones are segmented using different methods including interactive threshold segmentation(ITS), OTSU threshold segmentation(OTS-TS), global threshold segmentation(GTS), maximum entropy threshold segmentation(METS) and cross entropy threshold segmentation(CETS).The artificially segmented images serve as the criteria for evaluating the accuracy of segmentations made with above-going techniques.The segmented images are processed with DICE, RVD and VOE standard evaluation algorithms to obtain the evaluation indices of each algorithm.The algorithm with the highest segmentation accuracy obtained through data analysis is then verified with coal and refuse.As evidenced by test result, the use of the METS algorithm can produce the best result and the refuse can be successfully indetified and extracted while the CETS algorithm is the least effective one among the others; in terms of accuracy and the effects that can be obtained, the ITS algorithm is only inferior to the METS algorithm while the OTSU-TS and GTS algorithms stand next only to the ITS algorithm and is better than the CES algorithm; and the results of threshold segmentation of coal and refuse are respectively consistent basically with that of each mineral.The study result provides a good coal pretreatment condition for subsequent intelligent coal cleaning operations.