Tumor necrosis factor (TNF), in spite of its name, is known to contribute to pathogenesis of several diseases, including rheumatoid arthritis (RA). TNF blockade is widely used as therapy that was already applied to millions of patients worldwide. On the other hand, experiments with TNF-deficient mice clearly showed that TNF evolved as protective cytokine with non-redundant beneficial functions. Using conditional gene targeting, we are dissecting distinct physiological functions of TNF produced by various cell types in the context of live animal with the idea that pathogenic functions may be restricted to only some cellular sources. If true, such sources should be targeted during therapy allowing other cells to produce protective cytokine with beneficial functions. In collagen-induced arthritis (CIA), a mouse model for RA, TNF produced by myeloid cells (M-TNF) and B cells (B-TNF) both contribute to disease, whereas T cell derived – TNF (T-TNF) is protective. Reactivation of Tb is known side effect of anti-TNF therapy. Interestingly, in experimental tuberculosis infection, mice without T-TNF die, while TNF from other sources appears functionally redundant. These genetic findings suggest the design of future TNF blockers that may be superior to the currently used drugs.