How　to　find　what　a　user　wants　in　tremendous　amount　of　Web　information　is　a　great　challenge　to　web　search　engine.　By　focusing　downloading　web　pages　on　a　given　domain,　focused　crawlers　can　save　a　great　deal　of　works　and　improve　the　quality　of　the　information　they　provide.　We　put　forward　a　method　of　focused　crawling-MatchLink.　It　uses　document　vector　model　to　evaluate　topic　relevance　of　the　anchor　and　uses　Naive　Bayes　algorithm　and　multilayer　classification　method　to　compute　the　topic　relevance　of　the　web　page　containing　the　anchor.　According　to　these　two　relevancies,　topic　relevant　web　pages　have　prior　claim　to　be　downloaded.　Experiment　shows　that　the　result　is　better　than　BestFirst　and　BreadthFirst.