In a prospective study of 7735 middle-aged 7 British men, 504 of whom died in a follow-up period of 7.5 years, there was a U-shaped relationship between alcohol intake and total mortality and an inverse relationship with cardiovascular mortality, even after adjustment for age, cigarette smoking, and social class. These mortality patterns were seen in all smoking categories (with ex-smoking non-drinkers having the highest mortality) and were observed in manual but not in non-manual workers. The alcohol-mortality relationships (total and cardiovascular) were present only in men with cardiovascular or cardiovascular-related doctor-diagnosed illnesses at initial examination. The data suggest that the observed alcohol-mortality relationships are produced by pre-existing disease and by the movement of men with such disease into non-drinking or occasional-drinking categories. The concept of a "protective" effect of drinking on mortality, ignoring the dynamic relationship between ill-health and drinking behaviour, is likely to be ill founded.