Diagnostic peritoneal lavage was carried out in 79 patients with acute pancreatitis, at a mean time of 7 h after admission to hospital. The presence of more than 10 ml of free peritoneal fluid, brown-coloured free fluid or mid-straw-coloured lavage fluid was the criterion used for the prediction of a severe attack by lavage. Prior to lavage the attack was assessed as mild or severe by the clinician and reassessed by him at 24 and 48 h. All attacks were finally classified as mild or severe by means of a simple clinical grading scheme. There were 61 mild and 18 severe attacks (including 4 deaths). Initial clinical assessment correctly predicted only 39 per cent of the severe attacks compared with a 72 per cent succwss rate for diagnostic lavage. All the mild attacks were correctly predicted by clinical assessment but lavage was wrong in 3 out of 61 cases (95 per cent success rate). By 48 h after admission clinical assessment was comparable to systems using multiple criteria in its predictive value, success rates being 83 per cent and 82 per cent respectively for prediction of severe attacks. We did not find either the presence of methaemalbumin in the serum or the presence of hypocalcaemia to be of additional value. Diagnostic peritoneal lavage was the most accurate early guide to severity, and correctly predicted all patients who developed shock or died.