A small and inexpensive computer was programmed to assist patients in making decisions regarding insulin delivery using constant subcutaneous infusion systems. Insulin dosage was based on gender, pre- and postprandial blood glucose, between-meal blood glucose, patient weight, time of day, and when appropriate, the carbohydrate content of food ingested. The system was self-adjusting based on postprandial and fasting blood glucose levels. A developmental phase in which the computer program was refined was undertaken with five highly trained type I patients using an insulin infusion pump. Then, based on the suggestions made by these patients, a final program was used by these same five patients for 1 mo. Computer-assisted insulin delivery resulted in lower mean blood glucose (162 versus 130 mg/dl) and hemoglobin A1c (7.2% versus 5.8%) values when compared with precomputer values. In addition, there was a significant increase in the frequency of blood glucose testing during the computer-assisted periods in that patients monitored their blood glucose 4.9 times per day during the physician-alone period whereas a mean of 7.5 glucose tests were performed during the computer-assisted periods. Patient response to the concept was overwhelmingly favorable. These studies demonstrate that computer-assisted insulin-delivery decision making is feasible, acceptable to patients already accustomed to pump use, safe, effective, and may provide a savings in terms of professional time.