Abstract: Pattern recognition method has been applied to earthquake prediction recently.Using one of the unsupervised classification methods--clustering or cluster analysismethod, we studied the earthquake, data of the North China and Ning-Xia regions. It has been shown that the result of grouping for maximum homogeneity is better than hierarchical clustering. This result corresponds to the alternatively high and low seismic activities, and it effectively simulates mathematically the phenomena of seismic activity in groups. Analysis of error and test of prediction indicates steadiness and small error.