As a screening test for Cushing's syndrome, the evaluation of late-night cortisol levels is indispensable. We evaluated the usefulness and accuracy of plasma, urinary, and salivary cortisol levels measured late at night for the diagnosis of Cushing's syndrome. High cortisol levels (> 5 microg/dL) during the night are indicative of Cushing's syndrome, although night plasma cortisol levels are not readily reproducible because of the stressful situation. There was no correlation between plasma and urinary cortisol levels late at night, and late-night urinary cortisol levels provided weak information for the diagnosis of Cushing's syndrome. By contrast, late-night plasma and salivary cortisol levels showed a positive correlation, and salivary cortisol sampling was found to be useful for the diagnosis of Cushing's syndrome, because more than 0.4 microg/dL of late-night salivary cortisol levels gave a sensitivity of 86% and a specificity of 100% in our hospital. This method is also useful for the diagnosis of early or mild stage Cushing's syndrome, so-called subclinical Cushing's syndrome. Inherent differences between assays make it difficult to define optimal diagnostic criteria. However, the relative levels of salivary cortisol ratio, which is presented as a relative level, compared with the mean levels of healthy subjects in each institute, is useful for the screening of Cushing's syndrome as the cut-off level of 1.5 shows both high sensitivity and specificity in subclinical and overt Cushing's syndrome. Late-night salivary cortisol measurement is therefore a primary method of choice in the screening of patients suspected of having Cushing's syndrome.