The purpose of this study was to determine whether meeting historical criteria for unsuspected Wernicke's encephalopathy (WE), largely under-diagnosed in vivo, explains why some alcoholics have severe neuropsychological deficits, whereas others, with a similar drinking history, exhibit preserved performance. Demographic, clinical, alcohol related, and neuropsychological measures were collected in 56 abstinent alcoholics and 38 non-alcohol-dependent volunteers. Alcoholics were classified using the clinical criteria established by Caine et al (1997) and validated in their neuropathological study of alcoholic cases. Our alcoholics who met a single criterion were considered 'at risk for WE' and those with two or more criteria with 'signs of WE'. Whole blood thiamine was also measured in 22 of the comparison group and 28 alcoholics. Of the alcoholics examined, 27% met no criteria, 57% were at risk for WE, and 16% had signs of WE. Neuropsychological performance of the alcoholic subgroups was graded, with those meeting zero criteria not differing from controls, those meeting one criterion presenting mild-to-moderate deficits on some of the functional domains, and those meeting two or more criteria having the most severe deficits on each of the domains examined. Thiamine levels were selectively related to memory performance in the alcoholics. Preclinical signs of WE can be diagnosed in vivo, enabling the identification of ostensibly 'uncomplicated' alcoholics who are at risk for neuropsychological complications. The graded effects in neuropsychological performance suggest that the presence of signs of WE explains, at least partially, the heterogeneity of alcoholism-related cognitive and motor deficits.