A rapid, specific, precise and reproducible radioimmunoassay (RIA) for measurement of thyroxine (T4) in unextracted human serum is described. The procedure compares the ability of standards and unknowns to compete with radioactive T4 for binding sites on a T4-binding antiserum produced in rabbits by immunization with human thyroglobulin (Tg). The assay is set up in the presence of 1) 150 μg of 8-anilino-1-naphthalene sulfonic acid (ANS) in all tubes, to mobilize T4 from its binding with thyroxine binding globulin (TBG), and 2) barbital buffer to inhibit binding of T4 to thyroxine binding prealbumin (TBPA). The standard curve of the assay is regularly linear from 0.3–10 ng, thereby allowing accurate measurement of 1.2–40 μg/100 ml of T4, in one attempt, using 25 μl of serum. The validity of the method in general and specifically the lack of interference by TBG in serum has been demonstrated by 1) superimposability of T4 standard curves in buffer and T4-free human serum, 2) observation of expected values of serum T4 in serial dilutions of hyperthyroid sera containing high concentration of T4, and 3) general comparability of T4 measurements by RIA and the Murphy-Pattee method (CPBA), if the latter is corrected for incompleteness of recovery of T4 in butanol-ethanol extracts (avrecovery 85.5%). In sera from some hyperthyroid patients, serurr T4 estimates by RIA exceed those by CPBA beyond the expected extraction loss of T4 in CPBA. This discrepancy in serum T4 estimates by these two methods declines when sera from the same patients are tested at a time when hyperthyroidism is under control during antithyroic drug therapy. It reappears when sera are testec during relapse of hyperthyroidism. A similar discrepancy is also observed in sera of some euthyroid subjects when tested following administratior of TSH. This discrepancy between RIA anc CPBA appears related to the presence in serurm of T4 covalently linked to serum protein(s).