Three major principles of learning a chord keyboarding task were investigated. Five subjects were taught 18 characters on a chord keyboard, then practiced improving their keying speed for about 60 hours. The first objective of the study was to observe long-term learning on a keyboarding task. The performance, in characters typed per minute, was recorded over the entire range of the experiment. Typing skill improved quickly in the beginning and then slowed, but performance had not reached a stable peak by the end of the experiment. The second objective of this study was to determine a function that describes performance progress from initial training to a high keying speed. Five functions were evaluated; a function which predicts the logarithm of the dependent variable (characters per minute) from the logarithm of the regressor variable provided a good fit to the actual data. The final form of the equation was CPM; = e^B₀T_i^B₁ where CPM_i; = performance in characters per minute on the i-th interval, T_i = the i-th interval of practice, and B₀ and B₁ are fitted coefficients. The second objective also considered the form that T_i (from the above equation) should take. Performance can be predicted from number of repetitions such as trials, or from amount of practice such as hours. Both trials and time were used as predictor variables and both provided equally accurate predictions of typing speed. Both also provided excellent fits in conjunction with the Log-Log equation. Thus, it appears the Log-Log function is fairly robust in predicting performance from different variables. The third objective was to investigate how many trials of performance are needed before the entire learning function can be reasonably determined. In this experiment, subjects practiced for an extended period of time (about 60 hours) so a fairly complete progression of performance could be gathered. Yet, it would be more convenient to collect data for only a few hours and deduce the ensuing performance of the subject. The coefficients of the Log-Log function were determined using only the first 25, 50, 100, 150, and 200 of the initial performance points (out of about 550 total actual data points). The mean squared error (MSE) was calculated for each of these fits and compared to the MSE of the fit using all points. It appears that at least 50 performance data points are required to reduce the error to a reasonably acceptable level. / Master of Science