The kinetics of the antigen-antibody reaction were examined systematically in four kinds of double-antibody radioimmunoassay (RIA). In all the RIAs, the dose-response curves obtained on delayed addition by 24 to 48 h of labeled antigens (curves B), were shifted downwards and to the left of those obtained on simultaneous addition of the reagents (curves A), resulting in improved sensitivity of the assay. On the contrary, the dose-response curves obtained on delayed addition of unlabeled antigens (curves C), were shifted upwards and to the right of curves A, resulting in reduced sensitivity. In human thyrotropin (hTSH) RIA, curves B and C approached curves A very little, even after 168 h of incubation. A similar phenomenon was observed with anti-hTSH antisera from five different sources at two incubation temperatures, and the dilution curves of 125I-labeled hTSH and unlabeled hTSH appeared to be parallel. Therefore, the phenomenon observed with hTSH RIA could not be attributed to the assay conditions or to peculiar properties of the reagents used. In insulin RIA, the reversibilities of the shifts of curves B and C were slight but comparable to those observed in hTSH RIA. In 1–3,5,3'-triiodothyronine RIA, curves B and C gradually approached curves A on prolonged incubation and curves B became nearly identical with curves A after 98 h of incubation. On the other hand, in α-fetoprotein (AFP) RIA, curves B and C did not approach curves A, even on prolonged incubation for up to 288 h. The “equilibrium affinity constants” of the antibodies were of the same order of magnitude, thus it is unlikely that differences in the constants can account for the differences in the reversibility of these RIAs. In APF RIA, a significant amount of the antigen-antibody complex was precipitated without second antibody after centrifugation at 3000 × g. These findings suggest that the extent of dissociation of the immune complexes depends on their size, which in turn is related to the molecular weight of the antigen.