To describe the occurrence of severe gastrointestinal bleeding in community-dwelling older persons and to examine whether disability is a risk factor for this life-threatening condition independent of other known predictors. Prospective cohort survey. Three communities of the Established Populations for Epidemiologic Studies of the Elderly (EPESE). 8205 persons age > or = 68 years. The hospital discharge diagnoses provided by the Medicare Provider Analysis and Review files and the death certificates were prospectively surveyed for 3 years. Those with at least 1 discharge diagnosis of gastrointestinal bleeding and who received a blood transfusion or died were identified as cases of severe gastrointestinal hemorrhage. Physical disability, cognitive function, smoking and alcohol intake habits, body mass index, blood pressure, chronic conditions, number of hospital admissions in past year and medications taken were assessed at baseline. The occurrence rate of severe gastrointestinal bleeding was 10.8 per 1000 person-years (241 events/22,277 person-years). In proportional hazards regression models, compared with no disability, > or = 1 disabilities in the Rosow-Breslau scale (RR = 2.1, 95% CI = 1.5-2.9), and > or = 1 ADLs limitations (RR = 3.1, 95% CI = 2.1-4.6) independently predicted gastrointestinal hemorrhage after adjusting for age, gender, body mass index, comorbidity, number of hospital admissions, blood pressure, intake of coumarin, corticosteroids, aspirin and other nonsteroidal anti-inflammatory drugs. In this prospective analysis, disability is an independent predictor of gastrointestinal hemorrhage. Further studies are needed to explain the mechanisms by which disability may cause gastrointestinal hemorrhage. Because physical disability is potentially modifiable, strategies to lower the risk of gastrointestinal bleeding should be evaluated.