Recommender Systems are widely used to achieve a constantly growing variety of services. Along with social networks, recommendation systems that take into account friendship or trust between users have emerged. In this article, we propose a recommendation algorithm that respects data privacy constraints in order to be implemented on P2P architecture. It does not need any global knowledge on the network, it limits data exchange to trusted relations and protects private data from being spread over the network. We present an evaluation of different recommendation algorithms, to compare our to centralized ones, such as those used in online marketing services. Several concerns are considered: specific accuracy measures are defined, as well as the amount of required knowledge on the overall dataset. We ran two series of simulation tests that are herein presented and discussed.