Regulatory CD4(+) T cells, enriched in the CD25 pool of healthy individuals, mediate natural tolerance and prevent autoimmune diseases. Despite their fundamental and potential clinical significance, regulatory T (T(R)) cells have not yet been incorporated in a coherent theory of the immune system. This article reviews experimental evidence and theoretical arguments supporting a model of T(R) cell dynamics, uncovering some of its most relevant biological implications. According to this model, the persistence and expansion of T(R) cell populations depend strictly on specific interactions they make with antigen-presenting cells (APCs) and conventional effector T (T(E)) cells. This three-partner crossregulation imposes that T(R) cells feed on the specific autoimmune activities they suppress, with implications ranging from their interactions with other cells to their repertoire selection in the periphery and in the thymus, and to the relationship between these cells and the innate immune system. These implications stem from the basic prediction that the peripheral dynamics sort the CD4(+) T-cell repertoire into two subsets: a less diverse set of small clones of autoreactive effector and regulatory cells that regulate each other's growth, and a more diverse set of barely autoreactive T(E) cell clones, whose expansion is limited only by APC availability. It is argued that such partitioning of the repertoire sets the ground for self-non-self discrimination.