Although laser welding has many advantages such as fast processing time and single-sided access, the requirement of tight part-to-part gap control has been a main obstacle to maintaining the quality of laser welding. Traditionally, various stochastic anomaly detection methods have been developed for on-line weld defect detection, so that physical signals during the welding process can be monitored and classified. In order to improve the accuracy of weld defect detection, in this research paper, plasma intensity, weld pool temperature and back reflection signals are monitored, and their nominal trends are estimated by PDF estimation methods. We then aggregate these information, based on Dempster-Shafer theory. The performance of the proposed method is compared to the commercially available solutions of PRECITEC’s LWMTM and Hotelling’s T² method that are widely used in the literature. The proposed method reveals better performance in terms of Type I and Type II errors.