1. The recognition of the role of insulin resistance in disease states and the recent development of new drugs that modify insulin-dependent metabolism has led to increased use of the euglycaemic hyperinsulinaemic clamp to measure in vivo insulin sensitivity, but several key aspects of the technique are poorly documented in the literature. 2. We have evaluated the reproducibility and intersubject variation of measurements of insulin sensitivity in groups of insulin-sensitive and insulin-resistant subjects and assessed the effects of hand warning on haemodynamic and metabolic responses. 3. Subjects participated in one of two protocols: (i) 18 healthy male volunteers and 18 patients with hypertension and glucose intolerance were clamped on two occasions, 1 week apart with measurements of insulin sensitivity (M) derived after 120 and 180 min of hyperinsulinaemia; and (ii) six healthy volunteers were clamped on one occasion with simultaneous sampling of antecubital and 'arterialized' (dorsal hand) venous blood for comparison of plasma glucose concentrations and oxygen saturation and a further six volunteers were clamped on two occasions with and without the use of hand warming. 4. Measurements of M derived after 120 min (M120) and 180 min (M180) of hyperinsulinaemia were reproducible: the coefficients of repeatability (mg/kg per min) of M120 and M180 were 1.0 and 0.9 for volunteers and 1.0 and 1.0 for the patient group, respectively. The intersubject variation in insulin stimulus was high: coefficients of variation for M180 were 22% for volunteers compared with 38% for the patient group. In volunteers compared with the patient group, hand warming significantly increased venous oxygen saturations (95 +/- 2 vs 79 +/- 18%, respectively) and glucose concentrations (5.2 +/- 0.2 vs 4.5 +/- 0.4 mmol/L, respectively) and measurements of M were significantly higher using arterialized compared with antecubital venous blood. However, local hand warming was associated with systemic vasodilatation: blood pressure decreased (e.g. 6 mmHg diastolic; P < 0.05) with a compensatory increase in heart rate (8 b.p.m.). 5. In conclusion, clamps of 120 and 180 min duration yielded measurements of M that were reproducible. The technique is much more robust when used in the context of a crossover design because of the significant (20-40%) intersubject variation in M, even among apparently homogeneous male volunteers. Hand warming effectively arterializes venous blood and gives significantly higher M values, but induces systemic vasodilation, which may confound measurements of M.