The single-compartment pool fraction model, when used with the hyperinsulinemic glucose clamp technique to measure rates of glucose turnover, sometimes underestimates true rates of glucose appearance (Ra) resulting in negative values for hepatic glucose output (HGO). We focused our attention on isotope discrimination and model error as possible explanations for this underestimation. We found no difference in [3-3H] glucose specific activity in samples obtained simultaneously from the femoral artery and vein (2,400 +/- 455 vs. 2,454 +/- 522 dpm/mg) in 6 men during a hyperinsulinemic euglycemic clamp study where insulin was infused at 40 mU.m-2.min-1 for 3 h; therefore, isotope discrimination did not occur. We compared the ability of a constant (0.6 microCi/min) vs. variable tracer infusion method (tracer added to the glucose infusate) to measure non-steady-state Ra during hyperinsulinemic clamp studies. Plasma specific activity fell during the constant tracer infusion studies but did not change from base line during the variable tracer infusion studies. By maintaining a constant plasma specific activity the variable tracer infusion method eliminates uncertainty about changes in glucose pool size. This overcame modeling error and more accurately measures non-steady-state Ra (P less than 0.001 by analysis of variance vs. constant infusion method). In conclusion, underestimation of Ra determined isotopically during hyperinsulinemic clamp studies is largely due to modeling error that can be overcome by use of the variable tracer infusion method. This method allows more accurate determination of Ra and HGO under non-steady-state conditions.