Firstly, a distinguishable condition is proposed for separating the features by linear classification hyper surface. Secondly, the paper analyses the properties of the feature linear distinguishable criterion based on support vector machines (SVMs). Finally, the efficiency rate of features are defined by the contribution to classes margin of each feature, and a feature selection algorithm is put forward based on the feature efficiency rate. As experimental results show, validated with the actually measuring data and UCI (University of California, Irvine) data, performance of the new feature selection method, such as classification capability and generalized capability are improved obviously in contrast to the classical Relief method.