In these days there are many webshops, multimedia services, and other systems that provide very large number of items to the users. In systems like these is very useful to implement recommendation or collaborative filtering to help the users find the items they might like. There are many existing algorithm in classic client-server architectures, but recommending items can be valuable in P2P file-sharing systems too. But we have to face special problems and requirements in distributed architectures. There is no central database; every node in the network is just connected to a few other nodes. There are also no ratings or metadata about the items and users, the only information is the presence or absence of a file in a users file list. In my thesis I tried to give recommendations based on minimal binary information in a simulated P2P network. First I studied existing Collaborative Filtering solutions, and selected those, which can also work in distributed systems. After this, the next step was to build a P2P overlay network to connect the users with similar interest. A simulator have been designed and implemented, and based on a DC++ file list database it was possible to efficiently build a network with interest-based connections. With the simulated P2P network a few former selected Collaborative Filtering algorithms (user and item based CF and association rule based recommendation) were tested. After the experiments it turned out, that the success of the recommendation doesn’t depend on the number of the neighbors but on the aggregated similarity of the neighbors. After the evaluation of the results we can say, that based on the nearest neighbors shared contents, it is possible to make good recommendation to the target users with only limited knowledge about the entire network.