Because accurate determination of glycosylated hemoglobin (GHb) is difficult and relatively expensive in comparison with the modest cost and ready availability for tests of fasting plasma glucose (FPG), we examined the reliability of repeated measurements of FPG and GHb in typical diabetic outpatients taken in the usual clinical setting. We determined FPG and GHb concurrently on three separate occasions spanning 4 wk in 41 patients with non-insulin-dependent diabetes mellitus (NIDDM) and, for contrast, 5 with insulin-dependent diabetes mellitus (IDDM). Most of the NIDDM subjects were obese, with initial FPG levels ranging from 93 to 355 mg/dl. The reliability of each test was estimated by calculating two measures: the intraclass correlation coefficient (rho I) and the coefficient of variation (CV) for the repeated test values. For NIDDM patients treated with diet or oral hypoglycemic agents (OHA), rho I for FPG, log(FPG), and GHb were very similar. For insulin-treated NIDDM patients, rho I for FPG was somewhat lower than the coefficient in other treatment groups, and the reliability of FPG by this measure did not match the reliability of GHb within the limits of statistical significance. By analyzing the CV of test values repeated within subject, the reliability of FPG did not differ from GHb in any of the NIDDM treatment groups. Although patients were recruited sequentially to minimize sample selection bias, caution must be exercised in the interpretation of the statistical analyses of reliability with either rho I or CV due to limitations imposed by small sample size.(ABSTRACT TRUNCATED AT 250 WORDS)