Abstract: A T-Shaped harbor recognition method is proposed in polarimetric synthetic aperture radar (SAR) images based on the geometrical characteristic of T-Shaped harbor. Accurate coastline is extracted using the region-based level set method of polarimetric SAR in this paper. Feature points of the extracted coastline are detected using merge and split algorithm of digital curve, and the regions of interest (ROI) are searched by the short distance characteristic between beginning point and end point. Then the line segments of ROI are recognized by chain codes method and T-Shaped harbors are recognized by detecting parallel line and perpendicular line structure of the ROI. The algorithm is tested using polarimetric SAR images acquired by TerraSAR over the Bekerley area. Experimental results demonstrate the effectiveness of the proposed algorithm that T-Shaped harbors are correctly recognized in large-scale scene.