Tracer methodology has been applied extensively to the estimation of endogenous glucose production (Ra) during euglycemic glucose clamps. The accuracy of this approach has been questioned due to the observation of significantly negative estimates for Ra when insulin levels are high. We performed hyperinsulinemic (300 microU/ml)-euglycemic glucose clamps for 180 min in normal dogs and compared the standard approach, an unlabeled exogenous glucose infusate (cold GINF protocol, n = 12), to a new approach in which a tracer (D-[3-3H]glucose) was added to the exogenous glucose used for clamping (hot GINF protocol, n = 10). Plasma glucose, insulin and glucagon concentrations, and glucose infusion rates were similar for the two protocols. Plasma glucose specific activity was 20 +/- 1% of basal (at 120-180 min) in the cold GINF studies, and 44 +/- 3 to 187 +/- 5% of basal in the hot GINF studies. With the one-compartment, fixed pool volume model of Steele, Ra for the cold GINF studies was -2.4 +/- 0.7 mg X min-1 X kg-1 at 25 min and remained significantly negative until 110 min (P less than .05). For the hot GINF studies, Ra was never significantly less than zero (P greater than .05) and was greater than in the cold GINF studies at 20-90 min (P less than .05). There was substantially less between-(78%) and within- (40%) experiment variation for the hot GINF studies compared with the cold GINF studies. An alternate approach (regression method) to the application of the one-compartment model, which allows for a variable and estimable effective distribution volume, yielded Ra estimates that were suppressed 60-100% from basal. In conclusion, the one-compartment, fixed pool volume model of glucose kinetics is inadequate for the estimation of Ra during euglycemic glucose clamps. Two new strategies for estimating Ra from the one-compartment model, the hot GINF protocol and the regression method calculation, yielded more accurate and physiologically plausible estimates of Ra than currently used methodology.