Glucose tolerance depends on a complex interaction among insulin secretion from the β-cells, clearance of the hormone, and the actions of insulin to accelerate glucose disappearance and inhibit endogenous glucose production. An additional factor, less well recognized, is the ability of glucose per se, independent of changes in insulin, to increase glucose uptake and suppress endogenous output (glucose effectiveness). These factors can be measured in the intact organism with physiologically based minimal models of glucose utilization and insulin kinetics. With the glucose minimal model, insulin sensitivity (SI) and glucose effectiveness (SG) are measured by computer analysis of the frequently sampled intravenous glucose tolerance test. The test involves intravenous injection of glucose followed by tolbutamide or insulin and frequent blood sampling. SI varied from a high of 7.6 × 10−4 min−1 · μU−1 · ml−1 in young Whites to 2.3 × 10−4 min−1 · μU−1 · ml−1 in obese nondiabetic subjects; in all of the nondiabetic subjects, SG was normal. In subjects with non-insulin-dependent diabetes mellitus (NIDDM), not only was SI reduced 90% below normal (0.61 ± 0.16 × 10−4 min−1 · μU−1 · ml−1), but in this group alone, SG was reduced (from 0.026 ± 0.008 to 0.014 ± 0.002 min−1); thus, defects in SI and SG are synergistic in causing glucose intolerance in NIDDM. One assumption of the minimal model is that the time delay in insulin action on glucose utilization in vivo is due to sluggish insulin transport across the capillary endothelium. This was tested by comparing insulin concentrations in plasma with those in lymph (representing interstitial fluid) during euglycemic-hyperinsulinemic glucose clamps. Lymph insulin was lower than plasma insulin at basal (12 vs. 18 μU/ml) and at steady state, indicating significant loss of insulin from the interstitial space, presumably due to cellular uptake of the insulin-receptor complex. Additionally, during clamps, lymph insulin changed more slowly than plasma insulin, but the rate of glucose utilization followed a time course identical with that of lymph ( r = .96) rather than plasma ( r = .71). Thus, lymph insulin, which may be reflective of interstitial fluid, is the signal to which insulin-sensitive tissues are responding. These studies support the concept that, at physiological insulin levels, the time for insulin to cross the capillary endothelium is the process that determines the rate of insulin action in vivo. In separate experiments, a similar intimate relationship was found between lymph insulin and glucose utilization estimated from the minimal model, supporting the accuracy of the minimal model as a mathematical representation of insulin action in vivo. Additional factors in glucose tolerance are insulin secretion and clearance. We proposed a model of insulin/C-peptide kinetics, derived from the original conception of Eaton and Polonsky, in which determination of C-peptide kinetics in each individual is unnecessary if insulin and C-peptide kinetics are modeled simultaneously. Prehepatic insulin secretion after glucose injection was calculated in healthy women; total insulin secretion was 22.2 nmol; first-phase insulin averaged 38% of total, but there was wide variation among healthy subjects. The ability to determine insulin secretion, insulin action, and glucose effectiveness from modeling allowed us to investigate their interaction. We propose that in healthy individuals, there is a balance between secretion and insulin action such that insulin secretion × insulin sensitivity = constant. Thus, with insulin resistance, it is proposed that a normal β-cell will increase its sensitivity to glucose appropriately, staving off impaired glucose tolerance. This concept is supported by data in healthy pregnant women, in whom the reciprocal relationship is shown to exist and impaired glucose tolerance is not observed despite substantial insulin resistance (S, reduced to 1.8 × 10−4 min−1 · μU−1 · ml−1). In subjects at risk for diabetes, e.g., HLA-identical siblings of insulin-dependent diabetic subjects and Pima Indians, insulin sensitivity × secretion < normal constant value. Additionally, Pima Indians with the lowest sensitivity/secretion product appear to be at highest risk for developing NIDDM. Finally, with simulation of the models, we examined the relative importance of individual and compound defects of SI, SG, and insulin secretion to glucose intolerance. Although no individual defect (≤80%) of these factors causes diabetic glucose tolerance ( K G < 1), compound defects are remarkably synergistic, with a combined SI/SG defect being the most severe ( K G = 0.60), and an SG defect being a requisite component for diabetic glucose tolerance.