BACKGROUND Guidelines have been offered on haemoglobin thresholds for blood transfusion in surgical patients. However, good evidence is lacking on the haemoglobin concentrations at which the risk of death or serious morbidity begins to rise and at which transfusion is indicated. METHODS A retrospective cohort study was performed in 1958 patients, 18 years and older, who underwent surgery and declined blood transfusion for religious reasons. The primary outcome was 30-day mortality and the secondary outcome was 30-day mortality or in-hospital 30-day morbidity. Cardiovascular disease was defined as a history of angina, myocardial infarction, congestive heart failure, or peripheral vascular disease. FINDINGS The 30-day mortality was 3.2% (95% CI 2.4-4.0). The mortality was 1.3% (0.8-2.0) in patients with preoperative haemoglobin 12 g/dL or greater and 33.3% (18.6-51.0) in patients with preoperative haemoglobin less than 6 g/dL. The increase in risk of death associated with low preoperative haemoglobin was more pronounced in patients with cardiovascular disease than in patients without (interaction p < 0.03). The effect of blood loss on mortality was larger in patients with low preoperative haemoglobin than in those with a higher preoperative haemoglobin (interaction p < 0.001). The results were similar in analyses of postoperative haemoglobin and 30-day mortality or in-hospital morbidity. INTERPRETATION A low preoperative haemoglobin or a substantial operative blood loss increases the risk of death or serious morbidity more in patients with cardiovascular disease than in those without. Decisions about transfusion should take account of cardiovascular status and operative blood loss as well as the haemoglobin concentration.