Abstract: In this paper, single feature selection is done for 94 featues of seismicity using K-W test and entropy method. Best 12 and 10 features are selected respectivity by these two methods, and both results are more coincident with each other. For samples discribed by a small amount of features selected, the rate of mis-recognition of clustering by Fisher's method is lower than that by all 94 features. This shows it is effective that a small amount of features replace all features. Thus it can not only save time and cut down expenses but also avoid some problems caused possiblely by more features.