BACKGROUND Upper gastrointestinal (UGI) bleeding is a relatively common and potentially fatal complication of coronary artery bypass graft (CABG) surgery. However, little is known of this problem, including its incidence, predisposing factors and safety of endoscopy in these patients. OBJECTIVE To document the incidence, site, predisposing factors and outcome of UGI bleeding following CABG surgery. Also, to assess the safety of UGI endoscopy in these patients. METHODS Retrospective study of UGI haemorrhage following CABG at one institution between 1976 and 1991. RESULTS Fifty-five of 10,573 patients (0.5%) suffered a major UGI haemorrhage (as defined by need for transfusion or presence of melaena or haematemesis associated with hypotension). Of 51 patients undergoing endoscopy or laparotomy, 42 (82%) bled from duodenal ulceration. Five patients bled from gastric ulcers and one each from oesophagitis and Mallory Weiss tear. Nine patients underwent endoscopic therapy, which initially arrested haemorrhage in eight patients. However, three patients rebled and required surgery. Eight patients underwent surgery as initial therapy, resulting in an overall surgical rate of 20%. One patient died due to multi system failure following surgery. There were no complications from endoscopy. Patients who bled were more likely to have received inotropic support post-operatively prior to the haemorrhage (p < 0.05) and tended to be older than controls (mean age 65.6 years vs 58.7 years, p < 0.01). Twenty-one of the patients (38%) who bled had a past history of peptic ulceration or dyspepsia compared with 9% of controls (p < 0.001). Seven (12.5%) had previously bled from peptic ulceration. Patients who bled were less likely to have received H2-receptor antagonists in the perioperative period than controls (4% vs 20%, p < 0.05). CONCLUSIONS Upper GI haemorrhage following CABG is relatively frequent. It is usually secondary to duodenal ulceration. Endoscopy is a safe procedure in this patient group. Mortality did not differ between index patients who suffered a UGI haemorrhage and controls undergoing CABG who did not bleed.