The prognostic value of three predictive indices was compared in 436 attacks of acute pancreatitis in 418 patients. The outcome of an attack was graded as uncomplicated (324), complicated (70), or fatal (41); 1 attack was excluded because of puncture of the caecum. The overall mortality rate was 9.4%. Clinical assessment on admission identified only 34% of patients whose attack was severe (ie, complicated or fatal). Multiple laboratory criteria and peritoneal lavage were more sensitive (61% and 53%, respectively) while retaining diagnostic accuracy (79% and 74%). An erroneous diagnosis of pancreatitis was corrected by the findings on lavage in 5 patients. The major advantage of peritoneal lavage over multiple laboratory criteria was the shorter delay between admission to the study and determination of severity (median 4 v 24 h), but visceral puncture during insertion of the cannula occurred in 2 of 253 patients (0.8%). When used together, the three indices correctly predicted 82% of the attacks with a severe outcome and identified all patients destined to die within 10 days of admission.