A visual guidance method for the welding robot based on the weldment recognition and pose estimation is proposed to implement intelligent and flexible welding and solve the problem of that the visual guidance is affected by the three-dimensional structure of the weldment itself shielding the weld seam. Firstly, a welding model library is constructed by weldment CAD models off-line. Then, the viewpoint feature histogram(VFH) features of the weldment point cloud are calculated online and compared with the welding model library to implement weldment recognition. The fast point feature histogram(FPFH) features are calculated to implement a two-step pose estimation combining the sample consensus initial alignment(SAC-IA) and the ICP(iterative closest point) to obtain weldment pose. And the result is optimized by hypothesis verification. Finally, the welding trajectory is generated using the welding information in the welding model library combined with the pose of the weldment to provide visual guidance for the welding robot. The experimental results show that the method can accurately recognize the weldment and estimate its pose by using the target's overall three-dimensional information, and guide the robot to implement intelligent and flexible welding operations.