Abstract: Discretizing continuous attributes in a rough set were researched. Based on the concept of super-cube, all attributes of the information table in data space were globally discretized. By the consistent correlation of condition attributes and decision attributes, important condition attributes were selected depending on their classifying ability in the rough set boundary zone, and furthermore, important breaking points were selected to discretize the information table on a single attribute locally with the iterative constraints of information entropy. Illustration and experimental results indicate that the algorithm combining the global and local discretization is effective and efficient.