Aims/hypothesis. Adequate comparison of the relative performance of insulin sensitivity tests is not yet available. We compared the discrimination of four insulin sensitivity tests, commonly used in vivo, across a range of glucose tolerance. Methods. Normal (n = 7), impaired glucose tolerant (n = 8) and Type II (non-insulin-dependent) diabetic subjects (n = 9) had in random order two tests from the following: frequently sampled insulin-modified intravenous glucose tolerance test (FSIVGTT-MinMod); homeostasis model assessment (HOMA) and 2-h continuous infusion of glucose with model assessment (CIGMA) with immunoreactive or specific insulin; short insulin tolerance tests (ITT). The discriminatory power of tests was assessed by the ratio of the within-subject standard deviation to the underlying between-subject standard deviation (discriminant ratio –DR). The degree to which tests measured the same variable was assessed by comparing rank correlation with the maximum expected correlation given the imprecision of the tests. The unbiased lines of equivalence taking into account the precision of tests were constructed. Results. Reciprocal fasting plasma insulin (FPI^–1), HOMA %S and 2-h CIGMA %S, had similar DRs with ITT being less informative. The FSIVGTT-MinMod analysis was able to assess 13 out of 24 subjects and had a performance similar to ITT. Using specific rather than immunoreactive insulin for HOMA-CIGMA did not improve the DR. Reciprocal fasting plasma insulin FPI^–1, HOMA %S, 2-h CIGMA %S and S_I FSIVGTT intercorrelated more than 90 % of the expected rank correlation given the imprecision of the tests, but ITT gave only limited correlation. Conclusion/interpretation. The HOMA-CIGMA test with immunoreactive insulin provides similar information in distinguishing insulin sensitivity between subjects with normal glucose tolerance, those with impaired glucose tolerance and those with Type II diabetes as does FSIVGTT, whereas ITT is less informative. [Diabetologia (1999) 42: 678–687]