Aims. To assess the criteria used to identify alcohol dependence in the general population. Design and setting. Two independent probability surveys of the US household population 18 years of age and older were analyzed: the 1994 National Telephone Survey (NTS-94), which interviewed 637 respondents, and the 1988 National Household Interview Survey (NHIS-88) which interviewed 43 809 respondents in their homes. Participants. The analyses of the NHIS-88 dataset focused on drinkers who consumed at least 12 drinks of alcohol in the 12 months prior to the survey interview ( N = 22, 102). The analyses of the NTS-94 dataset focused on drinkers who consumed at least one drink in the 12 months prior to the survey interview ( N = 637). Measurements. Criteria for DSM-IV alcohol dependence were operationalized using 15 items from a standardized questionnaire. Findings. Analyses suggested that normal drinking behavior can be misidentified as dependence criteria. Results for men who drank up to two drinks per day suggest that if the dependence criteria were invalid, reductions in the prevalence of specific indicators of alcohol dependence would range from 0.3% to 5.2%. Correcting for the misidentification of alcohol dependence diagnosis would reduce the overall prevalence of alcohol dependence by 0.5%. Up to 7% of the men could have been diagnosed as alcohol-dependent and could have provided invalid reports. Conclusions. The identification of alcohol dependence in general population samples must include careful probing of the nature of drinking-related behavior reported by respondents. This will decrease misidentification of dependence criteria, increasing the validity of dependence diagnosis in survey research.