Data reduction, which can simplify large scale data and not lose useful information, is an important topic of knowledge discovery, data clustering and classification. Aiming to solve the current problem that continuous attribute in algorithm of clustering or classification has to be discrete, a new algorithm of data reduction based on cloud model is put forward. By use of cloud model, this algorithm calculates each conditional attribute's importance to decision-making attribute(s), and obtains the reduction attributes by virtue of greedy algorithm. This new data reduction algorithm was verified by some experiments and was proved to be stable and efficient.