Nonsteroidal antiinflammatory drugs (NSAIDs) are associated with a high prevalence of gastrointestinal toxicity. Comparisons of the risks associated with individual NSAIDs are needed, but most clinical studies in this area are not ideally suited for use in meta-analyses. These difficulties can be overcome by using strict criteria for the inclusion of studies, and by reanalyzing previous data, updated by authors where possible. In doing so, this meta-analysis compared the relative risk of serious toxicity associated with 14 NSAIDs from 12 studies with the risks of ibuprofen. These results were supported by a novel "summary ranking" analysis, which was weighted to limit the influence of smaller studies. Wide variations in relative risk among NSAIDs were observed with piroxicam and azapropazone being the most toxic. Ibuprofen was associated with the least risk, probably because of its widespread use as a low-dose analgesic. Five studies provided comparative data on NSAIDs at "high" and "low" doses (as defined in the original reports), showing that the risk of toxicity was dose related. Furthermore, at full antiinflammatory doses, the risk associated with ibuprofen was similar to that of naproxen and diclofenac. These analyses show that NSAIDs vary more in toxicity than in efficacy. First-line therapy should be started with the lowest effective dose of a less toxic NSAID, moving to higher doses or a more toxic NSAID only if the clinical situation demands it. Newer NSAIDs, such as selective cyclooxygenase-2 inhibitors, may provide safer antiinflammatory therapy.