To tackle problems of human-computer interaction the traditional scope of humanmachine studies needs extending to include the complex cognitive skills of understanding, communication and problem solving. This extension requires a fusion of the conceptual and empirical tools of human factors with those of cognitive psychology. A methodological approach to this fusion is outlined as a background for three studies of structured human-.computer dialogue. The studies involved a task in which secret messages were decoded in a number of discrete steps corresponding to computer commands. Each “command” required two numeric arguments. The study investigated underlying variables using questionnaire techniques in addition to user performance in an interactive version of the task. Three factors concerning the order of arguments in a command string were investigated: the consistent positioning of a recurrent argument, the relationship between argument entry order and their order in natural language, and the relationship between argument entry order and the position of argument values on a VDU. In Study I software specialists were asked to design command structures for the task and to give reasons for their choices. In Study II naive subjects were asked to choose between telegrams in which alternative argument orders were expressed in terms of alternative word orders. In the interactive version of the task, used in Study III, positionally consistent systems were most readily learned, but this depended on having the recurrent argument in the first position. With positionally inconsistent systems there were reliable effects due to the position of the direct object of individual command verbs. © 1981, Academic Press Inc. (London) Limited. All rights reserved.