Abstract: This paper presents two algorithms of linear feature selection used for the study of strong earthquake zoning in Beijing area. As a result, the new feature after selection describes better the physical properties of the pattern points and the difference between safe points and dangerous points of earthquake occurrence. The projected graph of points obtained in a feature space of decreased dimension not only shows well the distributive structure of pattern points but also has obvious effect upon distinguishing dangerous points and safe thus providing useful information for earthquake prediction.