AIMS: Computing the number of alcohol-attributable deaths requires a series of hypotheses. Using French data for 2006, the potential biases are reviewed and the sensitivity of estimates to various hypotheses evaluated. METHODS: Self-reported alcohol consumption data were derived from large population-based surveys. The risks of occurrence of diseases associated with alcohol consumption and relative risks for all-cause mortality were obtained through literature searches. All-cause and cause-specific population alcohol-attributable fractions (PAAFs) were calculated. In order to account for potential under-reporting, the impact of adjustment on sales data was tested. The 2006 mortality data were restricted to people aged between 15 and 75 years. RESULTS: When alcohol consumption distribution was adjusted for sales data, the estimated number of alcohol-attributable deaths, the sum of the cause-specific estimates, was 20 255. Without adjustment, the estimate fell to 7158. Using an all-cause mortality approach, the adjusted number of alcohol-attributable deaths was 15 950, while the non-adjusted estimate was a negative number. Other methodological issues, such as computation based on risk estimates for all causes for 'all countries' or only 'European countries', also influenced the results, but to a lesser extent. DISCUSSION: The estimates of the number of alcohol-attributable deaths varied greatly, depending upon the hypothesis used. The most realistic and evidence-based estimate seems to be obtained by adjusting the consumption data for national alcohol sales, and by summing the cause-specific estimates. However, interpretation of the estimates must be cautious in view of their potentially large imprecision.