Abstract: The fractal features of Internet topology was studied at router level by analyzing the massive data authorized by CAIDA (cooperative association for Internet data analysis) Skitter project. According to the definition of k-core, the Internet topology was divided into different cores at router level. Analyzing the main characteristic quantities such as degree distribution, degree correlations and clustering coefficient of every k-core, it was found that the degree distribution, clustering and correlation structure of Internet topology are all kept on as the network is pruned inwards step by step. The fact implies that there is a global self-similarity at the router level of Internet topology, which is approaching inwards to the center of the network. Furthermore, any region in Internet, which is defined according to the network centrality, has such a structure that its properties are the same to the topology of the whole network. All of these prove that Internet topology has fractal features at router level.