Peers and data objects in the Hybrid Overlay Network (HON) are organized in a n-dimensional feature space. As the dimensionality increases, peers and data objects become sparse and the distance measures become increasingly meaningless which leads to serious problems affecting HON performance. In this paper we propose a distributed feature selection technique reduce the dimensionaliy in HON. We study in our simulations the impact of the proposed feature selection technique on query results quality and show that it achieves high recall and precision.