Usability of an expert system is dependent upon the relationship between a human user and the expert system interface. The interface is defined as any combination of equipment with which the user and expert system communicate. Within the context of this research, the interface is considered to be the expert system text and graphics appearing on display hardware. This type of interface is known as an advisor. A state transition model is used to represent human-advisor interaction. The model provides a mechanism by which to collect objective human performance data. In addition, it is used to specify human-advisor interaction metrics. To test the state transition model, an expert system, Function Diagnostic, was developed. Function Diagnostic determines mathematical expressions for the graphical representation of selected piecewise linear and polynomial function. An experiment was performed in which subjects used Function Diagnostic to solve problems. Each problem was associated with one of three levels of difficulty: easy, moderate, and hard. The subject population consisted of 36 engineering students. Two subjective measures were recorded: (1) user confidence in solutions reached by Function Diagnostic and (2) user perception of problem difficulty. Objective measures associated with user errors and problem solving skills were also recorded. Expert system usability measures were derived from the human-advisor interaction metrics, and these measures are incorporated into a usability function. The usability function is a linear combination of (1) subjective measures and (2) the usability measures derived from the human-advisor interaction metrics. The function can be used to predict how a usability score will be changed when function variables are perturbed. / Ph. D.