An interesting problem in online social networks is the identification of user characteristics and the analysis of how these are reflected in the graph structure evolution. The basis of these studies are user similarity measures. In this paper, we approach user similarity from two angles. First, we propose a network similarity measure that considers only the graph structure and that, differently from existing techniques, takes into consideration also how two users are indirectly connected. Secondly, we propose a similarity measure based on user profile information, such to find semantic similarities between users. Moreover, since user profile data could be missing, we present a technique to infer them from profile items of the user contacts. We evaluate our similarity measures on Facebook and DBLP data.