Assessment of insulin sensitivity in subjects with non-insulin-dependent diabetes mellitus (NIDDM) is of paramount importance but intrinsically difficult. The standard (hereafter cold) minimal model, in conjunction with an insulin-modified protocol, has been recently proposed, but the estimates of insulin sensitivity showed poor precision (Saad et al. Diabetes 43: 1114-1121, 1994). We propose the tracer (hereafter hot) minimal model as a highly reliable method to estimate insulin sensitivity (SI*) and fractional glucose clearance (SG*), reflecting glucose disposal only, in NIDDM subjects. A [6,6- 2H2] glucose-labeled insulin-modified intravenous glucose tolerance test was performed in seven NIDDM subjects. In particular, SI* was 1.07 +/- 0.34 x10(-4)min(-1).microU-1.ml estimated with an average precision (mean coefficient of variation of 12%, range 4-22%), whereas the cold minimal model SI was 0.96 +/- 0.26 x 10(-4) min-1. microU-1.ml (mean coefficient of variation of 105%, range 3-353%). Another advantage of the hot indexes with respect to the cold indexes is their ability to reflect glucose and insulin effect on glucose disposal only, and not also on hepatic glucose production. Finally, we also studied by simulation the effect of glucose urinary loss on cold and hot minimal model indexes; only cold glucose effectiveness (SG) was significantly affected, resulting in a mean approximately 40% lower. The hot minimal model appears therefore more reliable than the cold model for assessing glucose tolerance in NIDDM subjects. In particular its ability to dissect disposal from production processes, coupled with the very good precision of the estimated metabolic indexes, supports the clinical use of this method in NIDDM subjects.