The aim of this study was to assess the association between alcohol intake and premature mortality (younger than 65 years) and to explore the effect of potential alcohol underreporting by heavy drinkers. We followed-up 20 272 university graduates. Four categories of alcohol intake were considered (abstainer, light, moderate and heavy consumption). Repeated measurements of alcohol intake and updated information on confounders were used in time-dependent Cox models. Potential underreporting of alcohol intake by some heavy drinkers (likely misclassified as light or moderate drinkers) was explicitly addressed in an attempt to correct potential underreporting by using indirect information. During 12 center dot 3 years of median follow-up (interquartile range: 6 center dot 8-15 center dot 0), 226 participants died before their 65th birthday. A higher risk of early mortality was found for the highest category of alcohol intake (>= 50 g/d) in comparison with abstention (multivariable-adjusted hazard ratio (HR) = 2 center dot 82, 95 % CI 1 center dot 38, 5 center dot 79). In analyses of alcohol as a continuous variable, the multivariable-adjusted HR was 1 center dot 17 (95 % CI 1 center dot 08, 1 center dot 26), for each 10 g/d of alcohol. This harmful linear association was present both in uncorrected models and in models corrected for potential underreporting. No significant inverse association between light or moderate alcohol intake and premature mortality was observed, even after correcting for potential misclassification. Alcohol intake exhibited a harmful linear dose-response association with premature mortality (<65 years) in this young and highly educated Mediterranean cohort. Our attempts to correct for potential misclassification did not substantially change these results.