The Modified Hierarchical classification Method is used to divide seismic active periods of Northern China in this paper. The seismic active periods divided are better when the distances between classes are increment of square sum of deviation and the distances between samples are Euclidean Distance than that when the distances between classes are any kind of method mentioned in the paper and the distances between samples are resemblance coefficient. The methods of predicting seismic active peroids and selecting features are proposed, and they are used as well. After features are selected by our method the classification results can be improved and the posterior prediction made are consistent with actual state basically.