Abstract: Duing to the high dimension and huge data of large database, data reduction plays more important roles in the knowledge discovery of database. Existing data reduction method for identify redundant and irrelevant attributes can be grouped into three type: enumeration search, heuristic search and randomized search,all have low efficient and may lost important information, and need improvement in order to satisfy the data mining requirement. Equivalence class partition based on the attribute set can identify the redundant and irrelevant attributes in essence, and can rapid and accurately compute the number of equivalence class partition, make data reduction more effective and data mining more efficient.