Hospital mortality after acute upper gastrointestinal haemorrhage varies widely. In a population-based, multi-centre, prospective survey of the management and outcome of unselected cases of acute upper gastrointestinal haemorrhage, we have assessed the effect of risk standardisation on this variation. We collected data from 74 acute hospitals in four health regions in the UK on patients aged 16 years and over who presented with acute upper gastrointestinal haemorrhage during 4 months in 1993 (3981 cases) and 3 months in 1994 (1584 cases). The overall mortality was 14.3% (798/5565). Crude mortality in individual hospitals ranged from 0% to 29%, and differed significantly from the overall rate in eight. Risk-standardised mortality ratios were calculated with a risk score derived from well-established risk factors. Only two hospitals had standardised mortality ratios significantly different from the reference value. When hospitals were ranked in order of increasing mortality, risk standardisation for age, shock, and comorbidity resulted in 21 of the 74 hospitals changing ranks by ten or more places. After further standardisation for diagnosis, endoscopic stigmata of recent haemorrhage, and rebleeding, 32 hospitals moved ten or more places from their original rank; one hospital moved 45 places. Risk standardisation to correct for variation in case mix results in apparently significant differences in mortality rates becoming non-significant. The current state of routine data collection does not allow for anything but the most basic case-mix adjustment to be made. Simple league tables of crude mortality are misleading in this disorder and cannot be regarded as a reflection of the quality of health care.