Road　information　is　an　important　geomatic　information　in　the　study　of　remote　sensing　images.　Its　automatic　extraction　from　remote　sensing　images　will　simplify　classification　and　measures　of　objects　in　the　images.　But　automatic　extraction　of　road　network　is　very　difficult.　In　this　paper　we　propose　a　semi-automatic　extraction　method.　Through　pattern　matching　and　processing　with　mathematical　morphology,　a　skeleton　of　road　network　can　be　obtained.　Experimental　results　show　that　the　method　is　quite　accurate　with　simplicity　and　rapidity　for　middle　or　high-resolution　remote　sensing　images.