One hundred and fifty-two patients were admitted to a single hospital with a diagnosis of acute pancreatitis during a 31-month period. Of these, 126 patients had contrast-enhanced abdominal computed tomography (CT) scans within 72 h of admission; 92 of these attacks were clinically mild, 34 were clinically severe. A single consultant radiologist reported the scans 'blind' and noted whether pancreatic enhancement was normal, increased or decreased, and whether there was loss of peripancreatic tissue planes. The maximum anteroposterior measurement of the pancreatic head and body were multiplied together to produce a 'pancreatic size index' (cm2). Significantly more patients with severe attacks had decreased pancreatic enhancement (79 versus 58 per cent, P = 0.01) and loss of peripancreatic tissue planes (82 versus 54 per cent, P = 0.006). The median (range) pancreatic size index for clinically severe attacks was 12.8 cm2 (3.0-52.5), and for mild attacks was 6.0 cm2 (1.1-23.4), P less than 0.0001. Modified Glasgow criteria had a sensitivity of 85 per cent and specificity of 79 per cent for clinically severe attacks. A pancreatic size index of greater than or equal to 10 cm2 had a sensitivity of 71 per cent and specificity of 77 per cent for clinically severe attacks. In conclusion, although there were highly significant differences between the clinically severe and mild groups with respect to pancreatic enhancement, peripancreatic tissue planes and pancreatic size indices, these CT criteria did not improve on modified Glasgow criteria for prediction of disease severity.