Insulin sensitivity is frequently assessed with the minimal model (MM) and the insulin-modified intravenous glucose tolerance test (MIVGTT). To ascertain the validity of this approach, the mechanisms by which the MM estimates glucose effectiveness and insulin sensitivity (SG and SI, respectively) from the data are analyzed theoretically. Published data and new labeled MIVGTTs in normal and non-insulin-dependent diabetic (NIDDM) subjects are used as reference data. One reason for the reported difficulty in estimating SI in NIDDM is the inadequacy of the MM to describe the data. SG is a biased estimate of the fractional glucose clearance (FGC) at basal insulin, and SI is a biased estimate of the average slope of the insulin concentration-FGC curve. The monocompartmental assumption and the role of glucose production in the MM are causes of bias. The bias is fairly constant across the spectrum of glucose tolerance (SG: approximately 150% overestimate; SI: approximately 30% underestimate). Also, SI is not expected to account satisfactorily for hepatic insulin sensitivity. Finally, by the use of SG and SI, FGC at a target insulin level (FGCMM) can be estimated. FGCMM agrees well with the analogous clamp index (difference 10%; correlation: r = 0.88, P < 0.002) and, in NIDDM, has a lower coefficient of variation than SI (57 vs. 82%). In conclusion, this analysis indicates that the MM is a sufficiently reliable method for the assessment of insulin sensitivity if it is used cautiously.