Visual sensing and adaptive control were combined in aluminum alloy pulse gas tungsten arc welding(GTAW) to control the penetration. The property of aluminum alloy GTAW was analyzed firstly. Stochastic system theory was used in the welding process, and the stochastic system model between the welding parameters and welding pool geometric parameters was established. Then real-time identification of model parameters was realized. Based on minimum variance adjustment, the adaptive controller for dominating weld formation was designed. The experiments show that adaptive control is an effective method for weld penetration in aluminum alloy GTAW.