Mangroves,　distributed　over　the　coastal　intertidal　zone,　have　a　composite　spectral　characteristic　of　water,　vegetation　and　soil.　The　extraction　of　mangroves　is　ineffective　using　conventional　remote　sensing　unsupervised,　supervised　classification　methods.　In　this　paper,　using　Landsat　ETM　+　as　data　source,　normalized　index　images　such　as　NDVI,　NDBI,　MNDWI　were　calculated.　By　analyzing　ground　-object　spectrum　characteristics,　the　decision　tree　classification　of　remote　sensing　was　made　to　study　the　method　of　extracting　mangrove.　The　results　show　that　the　knowledge　to　extract　mangrove　information　is　as　followed:　NDVI＞0　and　MNDWI＞0.13.　The　mangrove　knowledge　can　effectively　extract　mangrove.　Meanwhile,　the　knowledge　toextract　mangrove　information　also　verifies　that　mangrove　integrates　spectral　characteristics　of　water,　vegetation　and　soil　in　the　remote　sensing　images.