Automatic　building　extraction　from　high　spatial　resolution　remotely　sensed　imagery　can　accelerate　the　update　process　for　urban　basic　geographic　database.　One　problem　of　building　extraction　methods　is　the　difficulty　of　extracting　the　precise　building　contour.　This　article　proposes　an　approach　to　recognizing　and　extracting　buildings　from　high　resolution　remotely　sensed　imagery　based　on　shadows　and　graph-cut　segmentation.　Firstly,　shadows　were　detected　by　using　potential　histogram　function.　Then,　candidate　segmentation　objects　were　selected　from　the　result　of　graph-cut　segmentation　with　the　constraint　by　integrating　aspect　ratio　and　rectangularity.　At　last,　shadows　were　processed　with　open,　dilate　and　corrode　operations　respectively,　while　buildings　and　their　exact　boundaries　were　extracted　with　adjacency　between　processed　shadows　and　candidate　segmentation　objects.　For　verifying　the　validity　of　the　proposed　method,　six　sub-images　were　chosen　from　PLEIADES　images.　Experimental　results　show　that　the　average　precision　and　recall　of　the　proposed　method　are　92.31%　and　74.23%　respectively.　©　2016,　Chinese　Institute　of　Electronics.　All　right　reserved.