Estimation of size distribution by image analysis is a key issue in mineral engineering. However, only the surface information of ore piles can be captured, which is a headache problem in this field while only a few researchers pay attention to this problem. A new surface probability model was proposed for estimation of size distribution on a conveyor belt based on the Chavez Model in this investigation. This model was tested and verified to have smaller errors in single size fraction but have bigger errors in multiple size fractions. Several error trends were found and a correction factor was introduced to correct the higher errors. A series of linear equations were developed to calculate this specific correction factor according to Dm (average particle size) and the height of pile. Therefore, empirical probability can be estimated by the specific correction factor and calculated probability, and the surface information of ore piles can be converted into the global information of piles.