This paper discusses work on a model-based intelligent process controller for gas metal arc welding. Four sensors input to a neural network, which communicates to a reference model-based adaptive controller that controls process parameters. Reference model derivation and validation are discussed. The state of an arch weld is determined by the composition of the weld and base metal and the weld's thermomechanical history. The composition of the deposited weld metal depends primarily on the amount of filler metal dilution; heat input to the weld, comprising pre-heat and process heat, is the controlling factor in the thermal cycle. Thus, control of the arc welding process should focus on rational specification and in-process control of the heat and mass input to the weld. A control model has been developed in which the governing equations are solved for the process parameters as functions of the desired heat input (in terms of heat input unit weld length) and mass input (in terms of transverse reinforcement area) to the weld. The model includes resistive and arc heating of the electrode wire, characteristics of the welding power supply, and a volumetric heat balance on the electrode material, as well as latent and superheat of the electrode material. Extension of the model to include dynamics of individual droplet transfer events, based on incorporating a nonlinear, lumped parameter droplet analysis, is discussed. A major emphasis has been placed on computational simplicity; model solutions are required at the rate of about 10 Hz during welding. Finally, a process control scheme has been developed for the gas metal arc welding process using the above nonlinear model with a proportional-integral controller with adaptive coefficients to control the weld heat input and reinforcement area independently. Performance of the resulting control method is discussed. 10 refs., 5 figs.