Recently, triangular mesh models are widely used in product design and manufacturing. However, too many triangular patches are required for representing complex shapes accurately and it is also difficult to handle a meaningful set of triangular faces. To solve these problems, we propose a feature edge extraction method of triangular meshes based on segmentation using modified Watershed method and wide range normal evaluation. The proposed method robustly extracts feature edges from noisy mesh models, and enable to handle mesh models efficiently.