OBJECTIVE: Hepatocellular carcinoma (HCC) is one of the dreaded complications of cirrhosis. Although there are no randomized controlled studies showing improved survival with screening, patients with cirrhosis are screened for HCC. Little is known about the practice of HCC screening in the United States. Our aim was to describe the practice of HCC screening in patients with cirrhosis in the United States. METHODS: In March 1998, we mailed a standard questionnaire to 1021 physician members of the American Association of Study for Liver Diseases and the same questionnaire was re- sent to nonrespondents 4 weeks later. RESULTS: We received a response from 554 members (54%). After excluding those not involved in active adult patient care, 473 responses were eligible for analysis. Eighty-four percent of the respondents routinely screened patients with cirrhosis for HCC (screening respondents). Nearly half of the screening respondents limited the HCC screening to patients with high-risk etiologies such as hepatitis B or C or hemochromatosis. Although alpha-fetoprotein (99.7%) and ultrasound (93%) were the two most frequently used screening methods, a sizable proportion of the screening respondents (25%) used abdominal computed tomography for routine screening. On multivariate analysis, the following variables predicted screening for HCC by the respondents: seeing more than one new cirrhotic per week (odds ratio [OR]: 5.4, 95% confidence interval [CI]: 2.511.7); practicing for <10 yr (OR: 4.0, 95% CI: 1.2-13.4); an opinion that screening is cost-effective (OR: 6.4, 95% CI: 1.6-25); an opinion that screening prolongs survival (OR: 5.7, 95% CI: 1.8-17.9); and an opinion that not screening poses malpractice liability (OR: 9.3, 95% CI: 4.2-20.8). CONCLUSIONS: The majority of respondents routinely screen patients with cirrhosis for HCC. Approximately half of the screening respondents limit their screening to only patients with high-risk etiologies. On multivariate analysis, several variables predicted screening for HCC by the respondents.