The aim of this study was to test the validity of the CAGE questions as a measure of severe drinking in subjects at drop-in centers for the homeless, using biological markers of acute liver reaction to alcohol as the "gold standard." A sample of homeless men and women in Copenhagen were invited to participate in a study of health problems. Subjects were interviewed and blood samples were taken and screened for indicators of liver dysfunction (gamma-glutamyltransferase [gammaGT], mean corpuscular volume [MCV], alanine aminotransferase [ALAT], and alkaline phosphatase [Alpase]), and hepatitis C [HCV]. Scores on CAGE correlated strongly with years of heavy drinking (rho = 0.43, p < 0.001), and while years of drinking did not correlate with biomarkers after controlling for multiple hypothesis testing, CAGE correlated with gammaGT, Alpase and ALAT, but not MCV. The correlations held even among those without HCV, but subjects with HCV + and CAGE > 1 had quite extreme values on liver markers. Findings suggested that the CAGE was able to identify homeless drinkers whose drinking was significantly associated with increases in biomarkers associated with heavy drinking.