We compared serum TSH results determined in second and third generation assays in patients with thyroid disease and nonthyroidal illnesses (NTIs) to evaluate the usefulness of the more sensitive assay. We studied 19 subjects with untreated hyperthyroidism, 12 hyperthyroid subjects sampled at 4-week intervals after beginning carbimazole, 153 subjects receiving T4 replacement, and 300 hospital in-patients with a variety of NTIs. Serum TSH was measured, using a second generation immunometric method, together with free T4 and free T3. Samples with subnormal TSH (< 0.5 mU/L) were reassayed, using a more sensitive chemiluminescent immunometric method. Both assays revealed undetectable serum TSH levels in 18 of 19 overtly hyperthyroid patients. Undetectable TSH values (in both assays) were found in 30 of 33 patients with low serum TSH levels who were receiving treatment for hyperthyroidism, in association with normal thyroid hormone levels in 11. Undetectable TSH was evident in both patients receiving T4 and those with NTI, but use of the more sensitive assay led to a reduction in the number of subjects with undetectable TSH compared with the second generation results (T4-treated, 55 vs. 77 cases; NTI, 13 vs. 19 cases). There was a significant correlation between serum TSH and free T4 in the whole group on T4 (P < 0.001) and in those receiving T4 with low TSH (r = -0.33; P < 0.05); no significant correlation was evident in subjects with low serum TSH levels associated with NTI. An improvement in assay sensitivity led to a reduction in the number of patients being treated with T4 or with NTI in whom serum TSH was undetectable and, hence, an increase in those in whom overt hyperthyroidism could be excluded. Undetectable TSH results, even in a third generation assay, are not diagnostic of overt hyperthyroidism, but are also found in subjects with treated thyroid disease and NTI.