It is generally considered that T cells which are reactive with self-antigens are effectively eliminated by two processes: clonal deletion and the induction of T cell anergy. More recently, it has been shown that some potentially autoreactive T cells remain unactivated because the self-antigens for which they are specific are not presented on competent antigen-presenting cells. All these mechanisms of self-tolerance may be regarded as passive in the sense that the autoreactive cells are either deleted or are intrinsically non-responsive. If this view of self-tolerance is adopted, then one would predict that rendering animals relatively lymphopoenic should not give rise to autoimmune disease. This prediction is not verified by experiment. Rats rendered relatively lymphopoenic by adult thymectomy followed by repeated low dose gamma-irradiation develop a high incidence of autoimmune diabetes. Furthermore, it has been shown that the reconstitution of these rats with a specific subset of CD4+ T cells from syngeneic donors prevents the development of this disease. The protective cells have the CD45RClow phenotype, they are resistant to adult thymectomy and the majority of them appear to be non-activated in the donor rats. In contrast, the CD45RChigh CD4+ subset does not provide protection from diabetes. Instead, on injection into athymic rats, it gives rise to pathological changes in a variety of organs: stomach, pancreas, liver, thyroid and lung. In addition, the CD45RClow CD4+ subset prevents these manifestations of autoimmunity in these circumstances. Recently, we have shown that CD4+ CD8- thymocytes are a highly potent source of cells that have the ability to control autoimmune diabetes in rats. It appears that the thymus has three distinct functions: positive selection; negative selection; and the generation of a population of cells that seem specialized for the control of autoimmunity.