The application-layer network traffic classification is critical to network traffic control and manage-ment. In the traditional classification methods with supervised machine learning model，the same features are usedt0 G’lasslfy (hfferent app11G’atlons. It resuhs In that sOnle features G’an perform Well for sOnle app11G’atlons，but WOT’sefor Others. Thls paper prOpOSPS a netWOrk trafflG’ G’lasslflG’atlon frame hase(1 On SuhspaG’e }.1nSterlng-hase(1 MethO(1(SCM).SCM is utilized in feature selection extracts key features corresponding to each type of application and automatically eliminates irrelevant features. Then each type of application corresponds to its own signatures，which areapplied to classify the network traffic in the same overall feature space. The experimental results show that SCM extracts corresponding feature signatures effectively for each application. The classification accuracy of the proposedmethod is more than 93 %. And it identifies the new applications well.