Abstract: A new water-extraction method, which focuses on the characteristics of the Coastal water, was proposed in this paper. The method could extract water full-automatically, by using the scale transformation from pixel to object and from global to local, using the spectral features, special features, statistic features and uncertainty of Coastal water remote sensing information, and using the combination of geo-knowledge and data mining technology. Experimental result indicates that, the method could extract water in various types of coastal zones with better completeness, better continuity and higher accuracy than classic methods such as supervised classification and threshold segmentation.