Abstract Pretreatment with recombinant human granulocyte CSF (G-CSF) protected mice in two different models of septic shock. Intravenous injection of 250 micrograms/kg G-CSF to mice prevented lethality induced by 5 mg/kg LPS. Injection of 50 micrograms/kg G-CSF protected galactosamine-sensitized mice against LPS-induced hepatitis. In either case, this protection was accompanied by a suppression of LPS-induced serum TNF activity. In contrast, when galactosamine-sensitized mice were pretreated with 50 micrograms/kg murine recombinant granulocyte/macrophage CSF instead of G-CSF and subsequently challenged with LPS, serum TNF activity was significantly enhanced and mortality was increased. The suppressive effect of G-CSF on LPS-induced TNF production was also demonstrated in rats. In vivo, no TNF was detectable in the blood of LPS-treated rats, which had been pretreated with G-CSF. Ex vivo, alveolar macrophages, bone marrow macrophages, Kupffer cells, or peritoneal macrophages prepared from G-CSF-treated rats produced significantly less TNF upon stimulation with LPS than corresponding populations from control rats. However, when these macrophage populations were incubated with G-CSF in vitro, LPS-induced TNF production was unaffected. These data suggest that the G-CSF-mediated suppression of TNF production is not a direct effect of G-CSF on macrophages. To examine whether, independent of the protection against LPS, G-CSF treatment still activated neutrophils, it was demonstrated that granulocytes from G-CSF-treated rats were primed for PMA-induced oxidative burst and for ionophore/arachidonic acid-stimulated lipoxygenase product formation. The experiments of this study support the notion that G-CSF is a negative feedback signal for macrophage-derived TNF-alpha production during Gram-negative sepsis.