Abstract: Traditional intrusion detection methods are based on static data，and it cannot handle incremental data.However，new intrusion behaviors emerge endless.IDSs are required to learn the new types of intrusion.A novel scheme based on incremental principle component classifier is proposed.The experiments show that this approach can detect unknown intrusions more effectively，has a better performance in detection rate and false positive rate and the efficiency is improved at a certain extent even when the train data is incremental.