Understanding and modeling of the human welder's response to three-dimensional (3D) weld pool surface may help develop next generation intelligent welding machines and train welders faster. In this paper, human welder's adjustment on the welding current as a response to the 3D weld pool surface characterized by its width, length, and convexity is studied. An innovative vision system is used to real-time measure the specular 3D weld pool surface under strong arc in gas tungsten arc welding (GTAW). Experiments are designed to produce random changes in the welding speed resulting in fluctuations in the weld pool surface. Adaptive neuro-fuzzy inference system (ANFIS) is proposed to correlate the human welder's response to the 3D weld pool surface using three inputs including the weld pool width, length and convexity. The human welder's behavior is not only related to the 3D weld pool geometry but also relies on the welder's previous adjustment. In this sense, a four input ANFIS model adding the previous human welder's response as a model input is developed and compared with the fitted linear model. It is found that the proposed ANFIS model can derive a more accurate correlation between the human welder's responses and the weld pool geometry and help understand the nonlinear response of the human welder to 3D weld pool surfaces.