PURPOSE: To evaluate the use of thyroid function tests to monitor therapy in patients taking levothyroxine. DATA IDENTIFICATION: Studies published between 1966 and January 1990 were identified through computer searches and references of identified studies. STUDY SELECTION: Studies that included a careful description of the patients under study, well-defined interventions, and a meaningful criterion standard were emphasized. DATA EXTRACTION: Data from different studies were combined to estimate the probability of positive and negative test results in well-defined clinical situations. Informal methods of synthesizing evidence were used when combining data from different studies was not justified. RESULTS: The sensitive thyrotropin (TSH) test is the preferred method to monitor therapy because it agrees with physiologic measures of thyroid hormone effect. Among clinically euthyroid patients who take 100 to 150 micrograms/d of levothyroxine, the probability that the sensitive thyrotropin will be undetectable is close to 50%. These patients are most likely to benefit from testing. Patients who take over 250 micrograms/d are almost certain to have undetectable sensitive thyrotropin levels; in these patients, the dose may be lowered without testing. CONCLUSIONS: Adequate long-term studies are needed to determine the role of biochemical testing in monitoring therapy. Current evidence suggests that the sensitive thyrotropin test should be used to monitor therapy. Clinically euthyroid patients taking lower dosages of levothyroxine are most likely to benefit from testing.