Gangestad and Snyder (1985) have proposed two types of personality variables: continuous variables (characteristics possessed to some degree by all individuals) and class variables (characteristics distributed into discrete classes). This study examined whether alcoholism could be classified as a class variable. Because a class variable will exhibit a particular pattern among the covariance of its indicators, a pilot study was conducted to identify 10 indicators from the Drinking History Questionnaire, the MacAndrew Scale, the Comprehensive Drinker Profile, the Mortimer Filkens Test, and the Western Personality Inventory. The indicators were given to 125 male alcoholics and to 200 male nonalcoholics. Covariance among the items was plotted by level of response to the indicators. The covariance between each possible item pair was plotted for eight levels of responses to the indicators. The plot of the covariance among items by level of response to the indicators peaked toward the middle, indicative of the existence of a class variable. A high level of agreement between the simplest base rate estimation and the proportion of alcoholics in the sample provided support for the results. These findings suggest that alcoholism has a latent class variable underlying its development. A class variable, with its specific etiology, is much more likely to be genetic in origin than is a continuous variable, with its diffuse etiology. This finding has implications for the treatment and prevention of alcoholism. (NB)