A real-time monitoring system is developed for detecting abnormal conditions in robotic gas metal arc welding. The butt-joint test pieces with cut gap are used to intentionally introduce the step disturbance of welding conditions. During the welding process, the welding voltage and current signals are sampled and processed on-line to extract the characteristic information reflecting the process quality. After 1st statistical processing, it is found that seven statistical parameters (the mean, standard deviation, variance, and kurtosis of welding voltage; the mean, variance, and kurtosis of welding current) show variations during the step disturbance. Through 2nd statistical processing of the means of the welding voltage for subgroups of continuous measurement, the statistical control chart is obtained, and SPC (statistical process control) -based on-line identifying method is developed. Ten robotic welding experiments are conducted to verify the real-time monitoring system. It is found that the correct identification rate for normal and abnormal welding conditions are 100% and 95%, respectively.