We hypothesised that pharmacokinetic factors might go some way to explaining the risk of major gastrointestinal haemorrhage with non-steroidal anti-inflammatory drugs (NSAIDs), with bleeders exhibiting a reduced clearance of NSAIDs compared with non-bleeders and set out to investigate this. Fifty patients presenting to hospital with acute gastrointestinal bleeding while taking piroxicam, indomethacin, diclofenac or naproxen and age, sex, musculoskeletal disease and drug matched community dwelling controls, up to two for each index case, who had not bled were recruited. Clinical details including duration of therapy were recorded. Bleeders discontinued the implicated NSAID at presentation, controls at least five half-lives before the study. Bleeders were contacted by letter 1 month after discharge and invited to take part and were studied after a median delay of 5 months. Subjects received an oral dose of their respective NSAID and venous blood was sampled, over a period determined by the half-life of the NSAID. Plasma concentrations were determined by high performance liquid chromatography. The median length of treatment for the index patients was 1 year (range 2 weeks--28 years) and for the control patients 2 years (1 month--25 years), P<0.0005. There were no significant differences in peak plasma concentration, time to peak plasma concentration or area under the plasma concentration-time curve between bleeders or controls for any of the NSAIDs studied, apart from piroxicam Cmax being lower in bleeders at 2.07 mg l(-1) than in controls at 3.21 mg l(-1), mean difference (95% CI) -1.14 (-1.83 - -0.48), P<0.005. The data failed to support the hypothesis that reduced clearance of NSAIDs, which results in higher plasma concentrations, is a risk factor for acute gastrointestinal haemorrhage.