The sepsis syndrome has proved to be one of the most difficult clinical situations to replicate in animal models. The outcome of sepsis in patients depends not only on the pathogens, sizes of bacterial challenge and adequacy of treatment, but also on age, host defence mechanisms and individual response to septic shock and multiple organ failure. Novel immunomodulatory agents may control endotoxaemia or bacteraemia in animal models but they have not proved effective in septic patients. This article considers whether the principal problem lies in our interpretation of the data from animal studies, or whether such models are unsuitable for assessing new treatments for the sepsis syndrome.