BACKGROUND: Alcohol consumption has many different dimensions. For each potential medical or social outcome, different dimensions of consumption may have different relationships. However, these relationships are not independent of each other and this multidimensionality is often mishandled or not taken into consideration at all in current alcohol epidemiology. OBJECTIVE: To give recommendations on how to aggregate dimensions of alcohol consumption to predict social and medical consequences. METHODS: Based on a review of relevant papers, different statistical methods to deal with aggregating dimensions of alcohol consumption in predicting outcomes are compared and discussed. RESULTS: Regression approaches may be used to aggregate different dimensions of alcohol consumption to predict medical and social outcomes. However, the substantive interpretation of regression in general has to be taken into consideration. CONCLUSIONS: Future research in alcohol epidemiology should incorporate different dimensions of consumption and should analyze them by regression approaches either using the dummy variable approach or using suitable interaction terms.