In this study 50 patients with diabetes mellitus had gingival crevicular blood from periodontal probing collected in small plastic pipettes. The pipettes transferred the crevicular blood to a non-wipe glucose self-monitoring instrument. At the same time, finger-stick capillary blood measurements were analyzed in the same glucose self-monitoring instrument, and venous blood was collected for measurement in a laboratory glucose analyzer. Each laboratory measurement was corrected from a serum glucose value to a whole blood glucose value by a function of the patient's hematocrit. This corrected glucose value allowed direct comparison of the laboratory measurement to the intraoral and finger-stick whole blood measurements. The patient blood glucose concentrations ranged from 59 mg/dl to 366 mg/dl. The gingival crevicular blood exhibited a correlation of r = 0.975 (P < .0001) to the corrected laboratory standard measurement, with a mean prediction error (bias) of -4.11 mg/dl and a root mean square error (precision) of 17.43 mg/dl. The finger-stick blood had a correlation of r = 0.983 (P < .0001) to the corrected laboratory standard, with a mean prediction error of 4.65 mg/dl and a root mean square error of 14.48 mg/dl. The American Diabetic Association recommends that the prediction error of blood glucose monitoring devices fall within 15% of the laboratory standard. Using this criterion 92% of the gingival crevicular measurements and 90% of the finger-puncture measurements fell within 15% of the laboratory value.(ABSTRACT TRUNCATED AT 250 WORDS)