The state-of-the-art on arc welding research is considered, with unresolved questions and future directions highlighted. Both diagnostics and modelling are discussed. The focus is on the arc plasma, and its interactions with the electrode and workpiece, in tungsten-inert-gas and metal-inert-gas welding. Areas in which the need for further work is identified include development of techniques to measure current density distributions, calculation of the distribution of different gasses in the arc plasma (for example vapours of different metallic elements when welding alloys), computational methods for modelling metal transfer, and treatments of the sheath regions. It is shown that a thorough understanding of the arc is important in welding research and development. For example, reliable calculation of the heat flux to the workpiece requires the interactions between the arc and electrodes to be considered. Computational models of welding that take into account these interactions can already predict the shape and depth of the weld pool. Extensions of these methods would enable the determination of important properties of the welded metal, such as microstructure, residual stress and distortion, raising the possibility of the development of a "virtual manufacturing" capability.