The usefulness of measurements of serum alpha-fetoprotein elevation for diagnosis of the development of hepatocellular carcinoma was evaluated by a prospective study of 260 patients with cirrhosis. Hepatocellular carcinoma was found in 55 patients during the 5-yr follow-up, excluding 7 found to have hepatocellular carcinoma in the first 6 mo. The cumulative incidence of hepatocellular carcinoma was 26% in the 185 patients who had alpha-fetoprotein levels below 20 ng/ml at the time of entry and 46% in the 68 patients who had alpha-fetoprotein levels of 20 ng/ml or more but below 200 ng/ml. In 169 of the patients, serum levels of alpha-fetoprotein were assayed regularly for at least 2 yr. The incidence of hepatocellular carcinoma development in the 36 patients who had repeated transient increases in alpha-fetoprotein to above 100 ng/ml was 36%. This was significantly higher than the incidence in the 99 patients who had alpha-fetoprotein levels consistently below 20 ng/ml. Thus patients who had alpha-fetoprotein levels of 20 ng/ml or more, who had transient increases in alpha-fetoprotein or who had both should be treated as being in a super-high-risk group for hepatocellular carcinoma. Frequent and careful examination by ultrasonography of such patients is recommended.