Nineteen patients with insulin-dependent diabetes mellitus were evaluated for overall reliability of self-generated data from capillary blood glucose monitoring. For a period of 12 to 14 days, standard reflectance meters used by these subjects were replaced by meters internally modified with memory chips capable of storing all glucose readings by date and time. The subjects were not aware of this modification and were instructed to continue to test capillary blood glucose as they had been and to continue their practice of recording the meter readings in a logbook. To assess reliability of patient-generated data as recorded in the logbook, the addition, deletion, and alteration of test results were determined. A significantly lower (p less than 0.0001) mean blood glucose level was reported in the logbooks than recorded in the memory reflectance meters. Differences in logbooks and memory reflectance meters ranged from 0 to 109 mg/dl. Three fourths of the subjects had reported lower than actual mean blood glucose values. Under-reporting, or omission of memory reflectance meter readings, averaged 10 percent, whereas over-reporting or addition of phantom values averaged 40 percent. An average of 26 percent of the logbook entries were not identical to memory reflectance meter values determined at the corresponding time. Two thirds of the subjects had reported values in such a manner as to obscure hyper- and hypoglycemia, leading to misleading clinical impressions about the fluctuation in metabolic control. Previous glycemic control, patterns of logbook recording, or visits to the clinic were not found to be predictive of the reliability of patient self-monitoring regimens.