Experimental models of sepsis using endotoxin challenges, including studies with sensitized animals with D-galactosamine, have largely contributed to the basic rationale for innovative clinical trials in human septic shock, which have, to date, failed. The ability of these models to reproduce human disease has been highly discussed. We report here that the widely used D-galactosamine/LPS model does not account for septic shock. Treatment with YVAD-CMK, a potent tetrapeptide inhibitor of caspases of the interleukin (IL)-1beta converting enzyme (ICE) family, protects from LPS-induced liver apoptosis and mortality in D-galactosamine-sensitized mice when administered either before or up to 2 h after the lethal challenge. This curative effect is related to complete inhibition of caspase-3 activity in the liver. However, YVAD-CMK does not affect LPS-induced release of IL-1beta and does not protect from a lethal dose of LPS in unsensitized mice. These experiments demonstrate the difference between these two widely recognized experimental models of sepsis. LPS toxicity in D-galactosamine-treated mice, leading to blocked gene transcription, results from tumor necrosis factor (TNF)-alpha-induced caspase-3-dependent liver injury, not from the systemic inflammatory response. These results provide evidence that inhibitors of the ICE caspase family can prevent or even overcome the ongoing hepatic injury induced by TNF-alpha during sepsis, ischemia-reperfusion, or severe hepatitis.