This paper presents a variety of methods founded in cognitive science for examining the behaviour of the human operator in process control. The waterbath task described by Crossman and Cooke (1974) and by Moray et al. (1986) is taken as a basic process control laboratory paradigm for applying these methods in an integrated approach. First, the waterbath task itself was analysed and a graphical representation developed to show relations between system variables (various valve and heater settings) and state variables (volume, temperature and flow rate). From this a representation of possible correct knowledge about the waterbath was developed. Second, a classification was made of the different states the waterbath system could be in with respect to an operator's goal state. A state-action diagram identified the correct control action or actions to travel from each state to the goal. This provided a basis for predicting the types of difficulties operators might have. Results of an experiment snowed that the need to coordinate several control actions to reach a goal was a source of difficulty. Subjects consistently misunderstood the relation between variables and performance data showed poorer performance when these variables had to be controlled. Verbal protocol data showed how often and how accurately subjects discussed particular aspects of the waterbath system. It was possible to piece together fragments of a subject's whole system of knowledge. Overall, the methods discussed contrast possible actions and knowledge with actual actions and knowledge. They represent a promising collection of data reduction techniques for studying the human operator in process control.