Thyroxine (T4), triiodothyronine (T3) and reverse triodothyronine (rT3) concentrations in human milk were measured by radioimmunoassay in 114 samples obtained from 1 week to 8 months postpartum. Several assay systems applied for the determination of serum thyroid hormone concentration were proved to be unsuitable for human milk, and the method of separating free and antibody-bound hormone by polyethylene glycol was also inappropriate for milk specimens, which tended to give a falsely high value. The binding of finity of T4 to milk was lower than that to serum protein, on which 8-anilino-1-naphthalene sulfonic acid showed no remarkable effect. In spite of the high sensitivity of 100 pg/tub in T4 assay system, no immunoassayable T4 was detected in all samples with or without ethanol extraction and trypsin hydrolysates of milk. In contrast, T3 was present in a measurable amount in most of the samples, the mean +/- SD value of which was 10 +/- 9 ng/100 ml, and those in colostrum were significantly higher than those in matured milk (P less than 0.01), whereas rT3 was not detectable in 76 samples tested. These results indicate that permeability of thyroid hormones through the mammary gland is different between T4 and T3 as well as in placental transport, and human milk can not be a source of thyroxine supply for the breast-fed infant.