Sixteen patients with type I diabetes were randomly assigned to two groups to evaluate the utility of computer-assisted insulin dosage decision-making. All patients used the same solid-phase reagent strip system for glucose measurement and the same pump. The standard group (n = 9) used standard algorithms for insulin adjustment, whereas the computer group (n = 7) relied on interactive instruction from a small, inexpensive (less than $100) computer. At the beginning of the study, there were no significant differences between groups in C-peptide level, hemoglobin A1c level, age, or duration of diabetes. Mean blood glucose level during the study for the computer group was 121 mg/dl (6.7 mM), which was significantly lower (p less than 0.01) than glucose levels charted by the standard group: 148 mg/dl (8.2 mM). Mean number of blood glucose values charted by the computer group (58 per week) was significantly (p less than 0.01) greater than the number charted by the standard group (51 per week). Hemoglobin A1c values at six weeks correlated with the mean number of blood glucose values charted per week of the study. There was no difference between groups in symptomatic hypoglycemic episodes. Computer-assisted insulin dose decision-making is feasible, safe, and effective in enabling persons with type I diabetes mellitus to achieve lower mean blood glucose values over a six-week period while initiating pump therapy.