A method for measuring the unbound fraction of plasma cortisol which is suitable for clinical use is described. It is shown that unless some precise experimental conditions are fulfilled, important errors may result when determining the unbound fraction of the hormone. The assay is practical (2 ml of plasma are necessary) and reproducible (coefficient of variation: 2.7% within the same assay and 3.1% in different assays). Unbound cortisol measured at 0800 h in 86 healthy individuals was 9.7 +/- 2.6% (SD) of total cortisol and 15.0 +/- 8.5 ng/ml of plasma (temperature: 37 C). No significant difference was found between men and women or according to age. In most cases, variations of unbound plasma cortisol were more important than the variations of total plasma cortisol. This explains why unbound cortisol was a better discriminator in some pathological conditions. In Cushing's syndrome unbound cortisol was found to be increased on the average 2.9 fold whereas total cortisol was only increased by 53%. Unbound cortisol was especially high (24.7 +/- 3.8% of total cortisol and 78 +/- 18.5 ng/ml of plasma) in Cushing's syndrome due to adrenal carcinoma. In adrenal insufficiency, unbound cortisol averaged 6% of total cortisol and 1.4 ng/ml of plasma.