Failling hematocrit values are traditionally used to observe the course of active bleeding, since hematocrit values usually reflect acute blood losses. However, evidence from the literature suggests that, after volume replacement, some degree of normovolemic hemodilution may be desirable and that return to normal hematocrit values is not necessarily the appropriate goal of transfusion therapy. The optimal hematocrit value was defined empirically by three methods in a series of 94 critically ill postoperative patients. First, the mortality rates of postoperative patients were lowest with hematocrit values between 27 and 33 per cent. Second, mortality rates were examined when both hematocrit values and the important cardiorespiratory variables were reduced; significantly increased mortalties occurred when hematocrit values were less than an average of 32 per cent. Finally, oxygen availability and oxygen consumption increased significantly after whole blood and packed red cell transfusions were given when hematocrit values were less than 32 per cent but not above 33 per cent. When accurate blood volume measurements are not available, hematocrit values of 32 per cent are optimal; when volume therapy is indicated, blood may be given with hematocrit values less than 32 per cent, crystalloids or colloids are preferred with hematocrit values greater than 32 per cent.