Abstract: Objectives: Observing the backscattered electron images of lunar soil polished sections, some lunar soil particles are composed of a single mineral, and some particles are composed of multiple minerals. Since the number of particles in one lunar soil polished section is on the order of 105, we developed a computer-based classification method of lunar soil particles. Methods: We determine the approximate range of different minerals in the particle by grayscale statistics, and then use the watershed algorithm to segment the minerals in the particle. Finally, these lunar soil particles are divided into three categories: single mineral particles, dual minerals particles, three and above minerals particles. Results: After counting all lunar soil particles with an area >25 μm2 in the three polished sections, the statistical results showed that the percentages of number for a single mineral, dual minerals, and three or more minerals are 15.3%, 19.2%, and 65.5%, and the percentages of mass are 59.1%, 31.1%, and 9.7%. The average diameters of three kinds of particles are 31.29μ m,19.09μ m,9.27μ m. Conclusions: This difference in size of three types of particles might be related to the particles composed of various minerals separating easily during space weathering.