The approach so far has been to lump together large numbers of critically ill septic patients. The advantage of this approach is that it facilitates the rapid progression of the clinical trial, often with the hope that a posthoc analysis of the data will identify the type(s) of patients who are most likely to benefit from the intervention. This approach appeals to the pharmaceutical industry because it may extend the range of indications of the product. The problem is that the interventions may be beneficial in some patients, whereas in others they may be ineffective or even harmful. As an example, corticosteroid therapy has been shown to be ineffec- tive in large trials of patients with the sepsis syndrome,7 but a smaller study, restricted to pa- tients with typhoid fever, did show a protective effect9 A trial of penicillin conducted today in a heterogenous septic population would probably give negative results. Posthoc data analysis may prove too difficult to identify specific indications so that real beneficial effects may go unrecognized. One may draw a parallel between the clinical trials on sepsis and the large trials on oxygen delivery in critically ill patients: Increasing oxygen delivery to supranormal values may be beneficial in some subsets of patients with tissue underperfusion, but harmful in those who are already fully resuscitated. Globally, such interventions may not influence mortality rates.8 The issue may not be the number of patients, but rather their character.