In a quality review of glucose monitors, we measured the inaccuracy and imprecision of 26 systems. In each case, measurements on at least 50 capillary specimens from diabetic patients were compared with results from capillary blood that had been deproteinized and assayed with hexokinase. We also tested the monitors with commercial control solutions. In patients' specimens having a mean blood glucose concentration of about 9 mmol/l, the bias of the 26 monitors ranged from -5.1 to +20.1% (median=+7.5%). Imprecision of the monitors with blood specimens gave coefficients of variation (CV) ranging from 4.5% to 22.8% (median=8.7%) at the mean glucose concentration. A control solution for the monitors gave a glucose concentration of 7.6-13.6 mmol /l (median=9.2 mmol/l) with CV that varied from 1.7 to 19.8% (median=4.7%). While the means and CV's of the control were significantly correlated with bias and imprecision of the blood specimens, much of the variance remained unexplained (for bias, r(2)=0.17; for imprecision, r(2)=0.43). We conclude that a common basis for calibration could remove a significant component of variation and that control solutions may give a false impression of analytical performance.