In order to eliminate incompleteness and limitation in statistics of multi-granulation rough set, a multi-granulation rough set model was proposed by probabilistic rough set theory. First, probabilistic rough set theory was combined with multi-granulation theory and came up with the generalized definition about multi-granulation probabilistic rough set model. Abiding by the rules that positive region and negative region keep invariable, the attribute reduction for the optimistic model was defined. Then based on minimum Bayes expected risk decision theory, two parameters of multi-granulation probabilistic rough sets model were computed. Finally, the optimistic multi-granulation probabilistic rough sets model was applied to a case study of target recognition.