Physicians often must make prognostic judgments for critically ill patients, but we know little about how well they can perform this task. We prospectively measured disagreements among different physicians' quantitative prognostic judgments for 269 sequential admissions to an ICU, and evaluated the accuracy and discriminating ability of judgments made by different types of physicians. Many (44.7%) patients provoked one or more disagreements of at least 20 percentage points among the three possible pairings of physicians. Many patients whom one physician thought were certain to survive did not inspire such certainty in another of their doctors. The critical care attendings thought the changes of survival were less than or equal to 80% for 23, and between 81% and 95% for another 46 of 85 patients whom the patients' own attendings thought were certain to survive. There were some differences in the accuracy of the different physicians' estimates. All doctors showed excellent overall discriminating ability, but discriminated less well for postoperative patients. Lack of consensus about how to judge prognosis for critically ill patients may make it difficult to decide whom to admit to ICUs. These results underscore the need for valid predictive models to aid in decision-making for critically ill patients.