Thyrotropin (TSH) concentrations were measured in 1580 hospitalized patients and 109 normal persons. Using the mean +/- 3 SD limits of the log values for the controls (0.35-6.7 milli-int. units/L), the proportion of abnormal TSH results in the hospitalized patients was 17.2%. TSH was undetectable (less than 0.1 milli-int. unit/L) in 3.1% of patients, suggesting hyperthyroidism, and high (greater than 20 milli-int. units/L) in 1.6%, suggesting hypothyroidism. On follow-up of 329 patients, 62% with abnormal TSH (less than 0.35 or greater than 6.7 milli-int. units/L) and 38% with normal TSH concentrations, only 24% of those with undetectable TSH had thyroid disease: 36% of them were being treated with glucocorticoids and 40% had nonthyroidal illness (NTI). Although half the patients with TSH greater than 20 milli-int. units/L had thyroid disease, 45% of patients had high TSH values associated with NTI. TSH concentrations usually returned towards normal when patients' therapy with glucocorticoids was discontinued or they recovered from NTI. TSH test sensitivity appeared good when the mean +/- 3 SD limits of the reference population were used, i.e., no cases of hyper- or hypothyroidism, as identified by free thyroxin index (FT4I), were missed. However, TSH test specificity was inferior to that of the FT4I test (90.7% vs 92.3%), although specificity could be improved to 97.0% if the wider TSH reference limits of 0.1 to 20 milli-int. units/L were used--limits considered pathological if applied to outpatients. Evidently, different reference intervals for TSH are needed for hospitalized and nonhospitalized patients. We conclude that a "sensitive TSH assay" is not a cost-effective thyroid screening test for hospitalized patients as compared with the FT4I.