An online anomaly detection algorithm based on incremental projective non-negative matrix factorization is proposed to detect the network anomaly real-timely and efficiently. Firstly, an incremental projective non-negative matrix factorization is given, which has the same expression with PCA, and is able to construct normal and abnormal subspace to detect network-wide anomalies online by Shewhart control chart. Theoretic analysis indicates that, the proposed algorithm computation is far smaller than NMF-NAD. In addition, traffic matrix datasets analyzing for internet and simulation results show that the network anomalies detection algorithms based on NMF(such as NMF-NAD and ODA-IPNMF) performs better than that based on PCA, and the proposed ODA-IPNMF has comparable network anomaly detection by NMF-NAD, which the ability to detect the network anomaly online.