The systemic inflammatory response syndrome and its sequelae, sepsis, severe sepsis, and septic shock, remain the leading cause of death amongst adults with critical illness in many developed societies (1, 2). Defining criteria for these categories were first introduced by a Consensus Conference in 1991, employing perturbations in commonly measured clinical and physiological criteria primarily to facilitate patient enrollment in trials of new therapeutic interventions (3). Furthermore, an associated rise in 28-day mortality demonstrated for each of these categories of 10%, 20%, 20%–40%, and 40%–60%, respectively, has lent credence to the clinical value of what has been termed “the sepsis syndromes” (4). Since their inception, the sepsis syndrome defining criteria have assumed totemic status as representing the clinical manifestation of the host response to infection. This attractive concept was not, however, the epidemiological paradigm that it might appear. Systemic inflammatory response syndrome criteria alone appear to be overly sensitive and nonspecific, and many randomized controlled trials of potential therapies conducted during the 1990s foundered at least partly through the recruitment of heterogeneous patient populations (5). Consequently, a conference was convened in 2001 to reappraise, enhance, and improve the 1991 definition (6). Whilst the 2001 criteria were also adopted as criteria for trial entry, no formal evaluation of whether they represented a significant and clinically relevant improvement upon the 1991 approach has been conducted. The retrospective, observational study conducted by Zhao and colleagues (7) published in this issue of Critical Care Medicine addresses this deficiency in comparing the 1991 and 2001 consensus criteria in a random sample of nearly 1,000 critically ill patients. The authors employed a method involving sensitivity, specificity, and the area under the receiver operating characteristic curve for the criteria comprising the two consensus definitions by comparing the number of patients within their population that met or did not meet each of these vs. the number who were or were not considered septic from an independent adjudication by three clinicians. The 1991 definition was confirmed as having a high sensitivity but low specificity; indeed, recognition of this by international authorities provoked the 2001 revision. Interestingly, applying the 2001 criteria to the same population slightly increased sensitivity but decreased specificity; precisely the deficiency in the 1991 criteria that participants in the later conference attempted to correct. However, other interesting factors emerged. First, the sensitivities of both definitions were lower if it was specified that the criteria should be met simultaneously within a 24-hr time window rather than within the total intensive care unit (ICU) length of stay. Secondly, fever, high white cell blood count (or immature white blood cell numbers), low Glasgow Coma Score, the presence of edema and/or positive fluid balance, high cardiac index, low Pao2/Fio2, and high levels of creatinine and lactate were significantly associated with sepsis using both the 1991 and 2001 definitions and independent adjudication. How robust are the data? First, the authors correctly emphasize that their patient sample was derived only from ICU admissions, and their conclusions cannot therefore be generalized to the ward or emergency room. Second, we need to accept that adjudication is an inexpert science; indeed, this was reflected by the statistic, which estimated interrater agreement at 60%—a reflection of clinicians’ difficulty in identifying an entity that they are trying to define rigorously. Third, the patient sample was drawn from a heterogeneous group managed in seven ICUs, albeit in a single academic medical center. Patients had a wide range of admission diagnoses spread over cardiovascular, gastrointestinal, respiratory, genitourinary, and urological pathologies. It is possible that the performance of any defining criteria may differ according to the patients’ predisposition or the nature of the insult. However, the authors reasonably contend that the ICUs from which the sample population was sourced represent the primary critical care facilities for that region, and its heterogenous nature is therefore less relevant. What conclusions can we draw from this study? First, the authors remind us that the 2001 conference was predicated partly on the need to identify patients with sepsis early rather than to facilitate trials. With this in mind, an ideal definition would help clinicians identify patients with high sensitivity and specificity and would be convenient to use. In that regard, their study demonstrated that across the entire ICU stay, the 2001 sepsis definition was not significantly better than that of the 1991 criteria, suggesting that using the more modern approach does not improve discriminatory power. They also highlight a deficiency in our application of the criteria and the time epoch within which we examine the criteria—should they be met simultaneously at a single time point, or within a 24-hr window, a shorter period, or even a longer period. The different rates of evolution of physiological changes make this a difficult question. The authors considered a 24-hr time window. Surprisingly, this performed less well than a window that matched a whole ICU admission—a period of time that may encompass multiple pathophysiological events other than episodes of sepsis. Second, the list of possible clinical and physiological manifestations of the systemic inflammatory responses included in the 2001 definition was shown in the current study to be probably too extensive and complicated. What, therefore, is the way forward? Zhao and colleagues may point the way by identifying positive predictors of sepsis. Thus, given the complexity of using the extended (2001) list, a possible solution would be to shorten this to the restricted set of criteria they identified. Those predictors listed above represent *See also p. 1700.