Mechanisms to extract the characteristics of network traffic play a significant role in traffic monitoring, offering helpful information for network management and control. In this paper, a method based on Random Matrix Theory (RMT) and Principal Components Analysis (PCA) is proposed for monitoring and analyzing large-scale traffic patterns in the Internet. Besides the analysis of the largest eigenvalue in RMT, useful information is also extracted from small eigenvalues by a method based on PCA. And then an appropriate approach is put forward to select some observation points on the base of the eigen analysis. Finally, some experiments about peer-to-peer traffic pattern recognition and backbone aggregate flow estimation are constructed. The simulation results show that using about 10% of nodes as observation points, our method can monitor and extract key information about Internet traffic patterns.