The three decades old vector model used in information retrieval proved to be qualitatively better than the widely used Boolean model. We propose an efficient implementation of vector model based on metric indexing, particularly on M-trees.