The reliability of patient-generated data from self-monitoring of blood glucose (SMBG) was studied in 14 patients with type I (insulin-dependent) diabetes mellitus treated by continuous subcutaneous insulin infusion (CSII) (7 women, 7 men). The reflectance meters (Glucometer I, Ames, Elkhart, IN) used by the patients were replaced for a period of 21 days by memory-reflectance meters; patients were unaware of the memory capacity of the new meters and were instructed to continue their practice of recording the meter readings in their logbook. This study compares the data recorded in the memory-reflectance meters with those reported in the logbook. The number of SMBG measurements was different in 11 patients (differences ranging from 2 to 66). Mean glycemia was similar (8.23 +/- 0.36 mM in logbook vs. 8.49 +/- 0.48 mM in memory-reflectance meters), but both the M value and mean amplitude of glycemic excursions (MAGE) index were lower when calculated from logbook data (38 +/- 5 vs. 48 +/- 7 mM, P less than .05 and 6.91 +/- 0.43 vs. 7.72 +/- 0.52 mM, respectively; P less than .05). Overreporting (addition of phantom values in logbook) and underreporting (omission of SMBG measurements from logbook) indexes were 19 +/- 7 and 12 +/- 3%, respectively. Precision (percent of identical values in logbook and in memory-reflectance meters at the corresponding time) was 77 +/- 6.8%. The number of SMBG measurements recorded in the memory-reflectance meter was negatively correlated with glycosylated hemoglobin [HbA1c; (r = -.85, P less than .001)], whereas overreporting was positively correlated with HbA1c (r = .76, P less than .01).