A method for decomposing the user cursor action component of human-computer interfaces into individual components based on the four categories: target size, target distance, target direction, and selection mode, was investigated. A primitive task set consisting of the Cartesian product of specific elements of the four categories listed above was proposed based on observation of user tasks and a cursor action benchmark task set was developed to measure a user's performance for each element of the set of primitive elements with a given cursor control device. An experiment was conducted to test the proposed primitive task set and associated benchmark task set as a predictor of performance for a set of representative graphics tasks. The predicted times and actual times were shown to have very strong correlations and the data were also shown to conform to Fitts' Law.

A description of the experiment, the data collected, and the analysis of these data are included.