One hundred seventy-four patients (179 admissions) were prospectively evaluated for the subsequent occurrence of upper gastrointestinal ("stress") bleeding after admission to a medical/respiratory intensive care unit. Evidence for either overt or occult gastrointestinal bleeding developed in 25 (14 percent). The group of bleeders had a higher mortality (64 percent versus 9 percent), duration of intensive care unit stay (median 14.2 versus 4.2 days), number of patients requiring mechanical ventilatory support (84 percent versus 26 percent), and duration of such support for those who required it (median 9.5 versus 4.2 days) than the group who did not bleed. In three patients, death was related to bleeding. Upon patients' admission to the intensive care unit, diagnoses of an acute respiratory illness (but not specifically chronic obstructive pulmonary disease), a malignancy, or sepsis were more common among those who subsequently bled. Of factors tested, a coagulopathy and the need for mechanical ventilation were most strongly associated with the risk of bleeding. Other factors did not add to the risk once these two were taken into account. Among patients receiving mechanical ventilation, the risk of overt bleeding was particularly low for those who required such support for less than five days (only 3 percent). It is concluded that (1) significant upper gastrointestinal bleeding occurring after medical intensive care unit admission is an uncommon event, and (2) prolonged mechanical ventilation and/or the presence of a coagulopathy are the most potent risk factors. Medical patients with either of the latter conditions are most likely to benefit from prophylaxis regimens against "stress"-induced upper gastrointestinal bleeding.