We have reviewed the clinical details and radiology of all patients with non-variceal gastrointestinal haemorrhage referred for angiography during an 8 year period. Forty-six patients with a history of overt bleeding had 49 technically satisfactory angiograms of which 29 were true positives but 16 were false negatives (sensitivity 64%, accuracy 63%). The accuracy of angiography in this group was much higher (92%) in those patients referred within 24 h of admission. Nine patients who had a history of iron-deficiency anaemia due to sub-acute blood loss had 9 angiograms. Five studies in this group were correct (two true negatives and three true positives) but there were four false positive studies. Angiography is an effective means of diagnosis in patients with acute gastrointestinal haemorrhage who have a negative endoscopy but best results will only be achieved when the study is performed with minimal delay following negative endoscopy. Angiography should also be considered in patients with continuing subacute blood loss when other investigations are negative.