The particle size analysis of rock particle image has a wide application in practice. However, there is a challenge to extract complex particle size information because of the mutual adhesion among the particles. Fuzzy c-means clustering(FCM) algorithm can segment image by multiple eigenvalues of image, but it's difficult to determine C the number of sample set of clustering and m the fuzzy weighting exponent. So this paper proposes the method consisted of adaptive fuzzy c-means clustering(AFCM) algorithm and the watershed algorithm which can effectively make up the weakness of AFCM algorithm in segmenting conglutinate objects. At the same time the method can prevent the over segmentation of the image caused by the watershed algorithm. The method, firstly, carries on median filter to the image, next, does the AFCM clustering to the image, and then carries on the watershed processing to the image, and gets the satisfactory segmentation image. Finally, counts the number of particles, calculates single particle pixel, and meanwhile make a statistic of the same pixel area and classify them. Upon completion of all of the above, the particle size of the rock particle image is analyzed commendably.