This paper introduces random sampling into traditional fuzzy rough methods and proposes a random sampling based statistical rough set model. The work focuses on how to bring random sampling into traditional rough set. First, random sampling is used to propose a concept of k-limit, which can dramatically reduce the amount of computation during the computing of lower approximation value. Then, statistical upper and lower approximation is formulated. By mathematical reasoning, sufficient theorem and proof are used to valid the reliability of new model. Finally, numerical experiments illustrate the efficiency of the proposed statistical rough sets.