The majority of cases identified by commonly used alcoholism criteria in general community surveys are mild ones, with few alcohol-related problems. We illustrate this using data on 2088 Australian male twins aged 28-89 (mean age = 42.7), including 1846 who reported more than minimal alcohol exposure when surveyed by telephone in 1992-3. Using latent class analysis of alcoholism symptoms reported by these twins, we identify five classes of respondent: those with no alcohol-related problems (49% of the sample, if we include those with minimal alcohol exposure); excessive drinkers (33%; and individuals with a history of mild (14% moderate (3% or severe problems (1% Symptom endorsement profiles associated with these different classes are illustrated. The two most severe classes constitute a substantial majority of those reporting alcoholism treatment, but a minority of those reporting alcohol-related auto accidents or injuries, recurrent hazardous alcohol use, or alcohol-related arrests. The excessive drinkers and persons with mild problems account for a much higher proportion of persons experiencing these outcomes, and thus represent an important group to study from a public health perspective. The use of latent class analysis to improve case detection using structured or semi-structured diagnostic instruments is also discussed.