The so-called "fever work-up" is time-consuming and costly. The authors examined the practices of medical house officers in obtaining blood cultures, an important part of this evaluation, as well as the ability of these physicians to predict bacteremia in febrile patients. They studied all 344 medical inpatients who experienced episodes of fever during two 30-day periods, as well as all 50 cases of bacteremia detected during these and two additional 30-day periods. House officers drew blood for culture within one day after the onset of fever in 52% of fever episodes. In 20% of these episodes only one set of cultures (representing one venipuncture) was obtained. House officers estimated the likelihood of bacteremia to be 20% or less in 15 of 40 bacteremic patients. They failed to obtain blood cultures promptly in 10% of bacteremic episodes and in 27% of episodes where the cause of fever was a nonbacteremic bacterial infection. They obtained prompt blood cultures in only a bare majority of febrile episodes, frequently underestimated the likelihood of bacteremia, and inadequately sampled blood for bacteremia. In this study, clinical judgment was not an adequate substitute for routinely obtaining blood cultures for febrile medical inpatients.