On the basis of low specificity, poor positive predictive value, and cost, there is at present no basis for routine assessment of thyroid function in acutely hospitalized patients, unless clinical features suggest the possibility of thyroid dysfunction, or a patient's background increases the likelihood of thyroid dysfunction. When used in severely ill patients, estimates of both thyroxine (T4) and thyrotropin (TSH) show a high prevalence of abnormal results, but lack specificity and have poor positive predictive value for true thyroid disease. When thyroid function is tested in the critically ill, the positive predictive value for true thyroid disease of both free T4 and TSH measurements could be improved by using wider reference intervals than for unselected populations. The knowledge of nonspecific disease-related abnormalities of triiodothyronine, T4, and TSH is not currently likely to yield useful prognostic information or to alter management for individual patients. Thyroid testing should be readily available for any acutely ill patient with any clinical features that suggest thyroid dysfunction, and for groups at increased risk of thyroid dysfunction. An initial abnormal result for either TSH or free T4 estimate should be followed by combined analysis of free T4 and TSH with the best available methodology. Diagnosis of thyroid dysfunction should be based on the T4-TSH relation rather than either value alone. Persistence of an apparent diagnostic abnormality should be confirmed before therapy is commenced.