Visualizing social networks presents challeges for both nodelink and adjacency matrix representations. Social networks are locally dense, which makes node-link displays unreadable. Yet, main analysis tasks require following paths, which is difficult on matrices. This article presents MatLink, a hybrid representation with links overlaid on the borders of a matrix and dynamic topological feedback as the pointer moves. We evaluated MatLink by an experiment comparing its readability, in term of errors and time, for social network-related tasks to the other conventional representations on graphs varying in size (small and medium) and density. It showed significant advantages for most tasks, especially path-related ones where standard matrices are weak.