Highly selective inhibitors of cyclooxygenase-2, such as rofecoxib, are hypothesized to have an improved gastrointestinal tolerability and safety profile compared with non-selective NSAIDs, which inhibit cyclooxygenase-1 and cyclooxygenase-2 non-selectively. This paper reviews data from randomized, double-blind, placebo-controlled studies which investigated the effects of rofecoxib and NSAIDs on the human gastrointestinal tract. In healthy subjects, rofecoxib 25 mg and 50 mg daily had no effect on gastric mucosal prostaglandin synthesis, whilst naproxen 1000 mg daily caused a 70% reduction. Therapeutic doses of rofecoxib 25 mg and 50 mg daily did not increase intestinal permeability or faecal blood loss in healthy subjects, whereas increases in both measures were seen with indometacin 150 mg or ibuprofen 2400 mg. A supra-therapeutic dose of rofecoxib (250 mg) given daily for 7 days did not induce an increase in gastroduodenal erosions in healthy subjects, whilst increased numbers of erosions were found in subjects given ibuprofen 2400 mg or aspirin 2600 mg. The endoscopic findings in healthy subjects were confirmed in two 6-month clinical studies involving 1516 patients with osteoarthritis; the incidences of ulcers following rofecoxib 25 mg or 50 mg daily were similar to placebo and less than ibuprofen 2400 mg. The advantage of rofecoxib over NSAIDs in these studies appears to translate into clinically relevant benefits; an analysis of 5435 patients with osteoarthritis found a significantly lower incidence of gastrointestinal perforations, ulcers and bleeds in patients taking rofecoxib compared with patients taking NSAIDs. Overall, the findings from these studies suggest that, as a result of cyclooxygenase-1 sparing, rofecoxib is significantly less gastrotoxic than non-selective NSAIDs, and may not differ from placebo.