Abstract: A novel algorithm based on beamlet transform for detecting weld seam edges was presented. Taking into account of some special characteristics of welding image processing, an orientation-thresholding step as well as a two-scan method to the standard beamlet-based line detection algorithm were introduced. Experiments are conducted to detect weld seam edges in noisy weld seam images to test the anti-noising performance of the algorithm. The result of experiments show that the algorithm is capable of directly extracting weld seam edges from highly noisy weld seam images without any pre-processing or post-processing steps, showing its high efficiency and prominent anti-noising performance. The two-scan method is particularly helpful when coping with low SNR weld seam images.