Abstract: In order to identify gap and edge defects in laser welding process, plasma visible light signal and infrared signal of molten pool during the process of stainless steel welding was monitored by a self-made optical fiber laser welding on-line monitoring system. And the signal processing methods of filtering and short time Fourier transform were used to analyze the optical signals with different defects and the variation law of time domain and frequency domain with welding defects was obtained. When a gap defect occurred, the time domain amplitude of the visible signal decreased from 1.5V to 0.2V without any components in the range of 1000Hz~4000Hz. The time domain amplitude of infrared signal decreased from 3.5V to 1.0V without any components in the range of 2000Hz~7000Hz. When an edge defect occurred, the time domain amplitude of the visible signal decreased from 2.0V to 0.5V and the time domain amplitude of infrared signal decreased from 4.0V to 0.5V without any components in the range of 0Hz~1000Hz for both signals. The results show that there is a certain correlation between the visible signal, the infrared signal and the laser welding state. The defects in the laser welding process can be effectively identified by the change of amplitude and frequency of the optical signal. The study has certain application value for actual production.