Antithrombin is shown to undergo a slow spontaneous conversion to its inactive latent conformation with readily discernible amounts present in plasma on incubation at 37 degrees C for 72 hours. More rapid conversion occurs on incubation of isolated antithrombin at 41 degrees C or 50 degrees C, but the appearance on electrophoresis of free latent antithrombin is preceded by the formation, in reciprocal proportions, of a new slow band. This slow component is shown to be a heterodimer of active and latent antithrombin. It can be isolated as a single stable band either by incubation of antithrombin or by mixing equimolar proportions of active and latent antithrombin under the same conditions that give overnight crystallization of the active/latent antithrombin heterodimer. Similarly, equimolar addition of latent antithrombin to plasma results electrophoretically in a quantitative shift to the slower heterodimer mobility. Clinically, the presence of latent antithrombin is potentially deleterious, because its linkage to form the heterodimer results in inactivation of the otherwise normal molecule linked to the latent antithrombin. In the case of alpha-antithrombin, because the dimer readily dissociates, there is only a 11% additive loss of activity, but with beta-antithrombin the dimer appears more stable, with the additive loss of activity from the normal beta component being 21%, increasing to 33% on stabilization of the dimer with heparin. This linked and selective loss of activity of beta-antithrombin provides an explanation for the unexpected severity of thrombotic episodes in heterozygotes with conformationally unstable antithrombins.