A Web graph refers to the graph that models the hyperlink relations between Web pages in the WWW, where a node represents a URL and an edge indicates a link between two URLs. A Web graph is normally a very huge graph. In the course of users' Web exploration, only part of the Web graph is displayed on the screen each time according to a user's current navigation focus. In this paper, we make use of a fast kernel-based algorithm that is able to cluster large graphs. The algorithm is implemented in an online visualization system of Web graphs. In the system, a Web crawler first generates the Web graph of web sites. The clustering algorithm then reduces the visual complexities of the large graph. In particular, it groups a set of highly connected nodes and their edges into a clustered graph with abstract nodes and edges. The experiments have demonstrated that the employed algorithm is able to cluster graphs.