This review updates previous systematic reviews to explore the overall levels of risk of serious upper gastrointestinal complications of treatment with non-steroidal anti-inflammatory drugs (NSAIDs) and to investigate the importance of dose in explaining variations in risk. Thirty-six eligible case control studies were published between 1985 and 2000 and involved 19,648 cases and 105,373 controls. Eight eligible cohort studies included around 400,000 exposed subjects and 1 million non-exposed controls. The pooled unadjusted odds ratio from case control studies (compared with non-use of NSAIDs) for serious gastrointestinal complications is 4.06 (CI95% 3.47, 4.75); the pooled odds ratio from cohort studies is 2.29 (CI95% 1.50, 3.51). Unadjusted odds ratios range from 1.81 for ibuprofen to 7.46 for piroxicam. These data require careful interpretation because the statistical power of analyses of individual NSAIDs is lower than that for the class as a whole. There is a rank order of risk for different NSAIDs at low doses but at higher doses the odds ratios tend to converge. Ibuprofen is associated with a lower risk of serious gastrointestinal complications than other NSAIDs; this advantage is probably lost at higher doses (>1800 mg/day).