The enormous popularity of Online Social Networks (OSNs) dramatically changed the way people communicate and interact with each other. Unfortunately, the centralized control employed by OSN providers, such as Facebook, have negative impacts as it leads to privacy concerns. In fact, OSN providers can directly access a big amount of information about users and infer further information when using data mining techniques (e.g., about the on line interactions among two users).In this paper we propose VirtualFriendship, i.e., a collaborative solution that allows OSN users to browse each others' "profile" and to exchange messages in the OSN, while the end-peers of the interactions remain anonymous with respect to the OSN provider. We demonstrate by a proof of knowledge implementation (VF-App) the feasibility of our solution. Furthermore, our analysis and thorough set of experiments also show that the overhead introduced by our solution is very limited, and we believe negligible from a user perspective.