To evaluate the hypothesis that, in terms of all-cause death, drinking alcohol 1-4 days per week is less harmful than daily (5-7 days/week) drinking of the same quantity of alcohol, a prospective cohort study using a self-administered questionnaire was conducted in Japan between 1990 and 2003 of 88,746 subjects (41,702 men and 47,044 women) aged 40-69 years at baseline. Among male regular drinkers consuming alcohol more than 1 day per week, light drinkers (<300 g/week) showed no increase in all-cause mortality irrespective of frequency of alcohol intake. Heavy drinkers (> or = 300 g/week), however, showed an increased risk of all-cause mortality among those who consumed alcohol 5-7 days per week, while no obvious increase was observed among those who consumed alcohol less than 4 days per week. Hazard ratios for drinkers who consumed alcohol 5-7 days per week were 1.29 (95% confidence interval: 1.12, 1.50) for 300-449 g per week and 1.55 (95% confidence interval: 1.32, 1.81) for > or = 450 g per week when compared with those for occasional drinkers who consumed alcohol 1-3 days per month. These findings support the Japanese social belief that "liver holidays," abstaining from alcohol for more than 2 days per week, are important for heavy drinkers.