Medical therapy is an attractive adjuvant to endoscopic treatment in upper gastrointestinal (UGI) bleeding. This review aims to assess the treatment effects of proton pump inhibitor (PPI) therapy in acute non-variceal UGI bleeding. Outcome measures evaluated were further bleeding, surgery, all-cause deaths, ulcer deaths and non-ulcer deaths. We searched MEDLINE (1966-2002) and EMBASE (1974-2002) using the terms 'gastrointestinal hemorrhage', 'peptic ulcer hemorrhage', 'proton pump inhibitor', 'omeprazole', 'pantoprazole', 'lansoprazole', 'rabeprazole' and 'esomeprazole'. The search was extended to the Cochrane controlled trials registry database, published abstracts from five international gastroenterology conferences, manufacturers of PPI, known contacts and bibliographies from each full-length published report. We included trials published in English and non-English languages. Eligible studies were randomized controlled trials that compared the treatment effects of PPI therapy with placebo or H2 receptor antagonists in patients with acute non-variceal UGI bleeding. Of the 175 articles screened, 26 controlled trials including 4670 subjects (2317 in treatment arm and 2353 in control arm) were analyzed. The methodology, population, intervention, and outcomes of each selected trial were evaluated using duplicate independent review. Disagreements were resolved by consensus. PPI therapy significantly reduced rates of further bleeding (odds ratio [OR], 0.48; 95% confidence interval [CI], 0.40-0.57) and surgery (OR, 0.61; 95% CI, 0.48-0.76). All-cause deaths were unaffected (OR, 1.02; 95% CI, 0.76-1.37). Ulcer deaths showed a significant reduction (OR, 0.58; 95% CI, 0.35-0.96), while non-ulcer deaths showed a significant increase (OR, 1.60; 95% CI, 1.06-2.41) in the PPI therapy group. Sensitivity analysis of 22 trials published in peer-reviewed journals, 10 trials with double-blind design and 19 trials with high quality score and 22 trials using omeprazole in the treatment group showed results similar to those seen in the analysis of all 26 trials, confirming the stability of the conclusions. Subgroup analysis revealed that summary outcome measures were not influenced by control group therapy (placebo vs H2 receptor antagonists) or the use of prior endoscopic treatment to achieve hemostasis (given vs not given). However, the summary treatment effects for further bleeding and need for surgery were significant in only those trials enrolling patients with peptic ulcers having high risk for rebleeding and not in those trials enrolling patients with all causes of UGI bleeding. The summary treatment effects for further bleeding and need for surgery were significant in trials using intravenous as well as oral PPI. However, summary OR for all-cause deaths and non-ulcer deaths in trials using intravenous PPI were higher in the treatment group and not in trials using oral PPI. This raised the possibility of intravenous PPI-therapy-associated non-ulcer deaths in high-risk patients. PPI therapy in acute non-variceal UGI bleeding reduced rates of further bleeding, surgery and deaths caused by ulcer complications. However, non-ulcer deaths were increased. The overall mortality was unaffected. PPI therapy is useful only in a selected group of patients with acute non-variceal UGI bleeding, namely those with peptic ulcers having endoscopic high-risk stigmata for rebleeding.