TC4 titanium alloy thin plates with the thickness of 1.5 mm are welded by Nd:YAG pulsed laser. A coaxial machine vision system that used high speed charge coupled device (CCD) is designed to acquire welded spot pictures. The welded spot pictures quality is significantly improved with the help of auxiliary illuminant. The welded spot edges are extracted by means of cellular neural network algorithm. The information of thin plate fenestration or insufficient depth of fusion can be acquired by analysis of welded spot pictures, and it is used as an input signal for closed-loop control. Experimental results demonstrat that this method can efficiently reduce thin plate fenestration or insufficient depth of fusion improve the welding quality.