Gastrointestinal bleeding (GIB) is known to be associated with use of nonselective nonaspirin nonsteroidal anti-inflammatory drugs (NNANSAIDs), but it is less clear whether and to what extent variation in GIB risk exists according to the characteristics of exposure. To assess the consistency of data on the relative and absolute risks of GIB among NNANSAID users, we reviewed epidemiologic studies conducted around the world for data on NNANSAID use categorized by dose, timing, and type of exposure. US mortality rates for GIB were obtained and used to estimate national rates of GIB death attributable to NNANSAID use in the 1990s. The average relative increase in risk of GIB among NNANSAID users was found to be fourfold or slightly higher, regardless of the demographic characteristics of the studied population. There were approximately twofold increases associated with over-the-counter doses and sixfold or higher increases at heavy prescription levels. Variation in risk exists by type of NNANSAID, although differences by preparation seem at least partly dose-related. GIB risk declines following cessation of NNANSAID use, with current rather than prior cumulative exposure the key determinant of risk. The baseline rate of GIB in all studied populations rises markedly with advancing age. Annual hospitalization rates for GIB exceed one per 1000 among the elderly, with the large majority of cases occurring among persons age 65 years and older. NNANSAID use may account for nearly 34% of all GIB cases in the United States, and may have resulted in over 32,000 GIB hospitalizations and 3200 GIB deaths per year in the 1990s.