OBJECTIVE To prospectively validate in patients with non-variceal upper gastrointestinal bleeding three risk scoring systems (the Baylor College scoring system, the Rockall's risk scoring system and the Cedars-Sinai Medical Centre predictive index) previously proposed to be predictive of rebleeding/death after upper gastrointestinal bleeding. METHODS We calculated values of the scores for 343 patients, who underwent endoscopy after non-variceal upper gastrointestinal haemorrhage during the years 1997-1999. We compared the observed outcomes with the ones expected upon the original series contributed by the authors. Discriminative ability was evaluated by calculating the area under the receiver operating characteristic curve. CONCLUSIONS Rockall's score accurately predicted rebleeding in low- and intermediate-risk categories (< 6), but not in high-risk patients. The rates of rebleeding were significantly higher than the ones predicted by the low-risk categories of either Cedars-Sinai index (< or = 2) or Baylor score (< or = 6). The predicted and the observed mortality was not significantly different throughout all the categories of Rockall's score, except for patients with a score of 4. All the scores had better discriminative ability for mortality than for rebleeding. The Rockall's score identifies a low-risk group of patients (Rockall's score < or = 2) for rebleeding and mortality.