Current treatment of human autoimmune diseases (AIDs) was developed empirically and relies mostly on non-selective suppression of the immune system. Traditional non-selective immunosuppressants such as corticosteroids, cyclophosphamide, and methotrexate and more novel means such as monoclonal antibodies to CD3, CD4, or CD25 do not discriminate between pathogenic and beneficial T cells. Importantly, the severe side effects seen with current therapies are related to the fact that these treatments not only suppress the pathogenic disease-inducing cells, but also cells influential in combating infections and killing malignant cells. Severe infections and malignancies are the inevitable result of non-selective immune suppression. Many of the novel forms of therapy of AID were developed in experimental animals, and their translation to the human disease was associated with the revelation of unexpected and sometimes catastrophic side effects. These surprises underscore the major differences between the relative simplicity of the experimental model and the complexity of the human disease. How can this current state of treatment of AID be improved? Which principles should guide us in the design of new treatments? This review attempts to offer a new look at these questions.