In this paper, a new method was proposed for retinal vessel detection. Most existed methods of vessel detection adopt pre-processing steps before vessel edge extraction, such as noise-filtering, enhancement, etc. These processing steps always change the vessel characteristics and result in inaccurate detection. Our proposed method is used to identify and eliminate such problems. After fundus region detection and illumination equalization, Canny operator is used to trace all edge using as candidate of vessel edge directly. Then line linking, cross-point removing, length filtering and curve filtering are adopted to find the true vessel edge. The preliminary results were showing that the proposed method has great potential for accurately vessel detection. Further, the method has a lower computation cost