To evaluate the relationship between serum cholesterol level and all-cause, coronary heart disease (CHD), and non-CHD mortality as a function of age. The data source was the biennial examination data from 1948 through 1980 for the 5209 men and women enrolled in the Framingham Heart Study. Age-specific analyses by the Cox proportional hazards regression model were performed of survival subsequent to ages 40, 50, 60, 70, and 80 years for all subjects enrolled and alive at each of the stated ages. Complementary models were studied that used high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, or total cholesterol level as predictors of survival subsequent to the examination at which lipoprotein subfractions were determined (1968) through 1973). The relationship between total cholesterol level and all-cause mortality was positive (ie, higher cholesterol level associated with higher mortality) at age 40 years, negative at age 80 years, and negligible at ages 50 to 70 years. The relationship with CHD mortality was significantly positive at ages 40, 50, and 60 years but attenuated with age until the relationship was positive, but not significant, at age 70 years and negative, but not significant, at age 80 years. Results for the relationship between low-density lipoprotein cholesterol and high-density lipoprotein cholesterol and mortality help explain these findings. Non-CHD mortality was significantly negatively related to cholesterol level for ages 50 years and above. The negative results in the oldest age group for all-cause and CHD morality appeared to be due to a negative relationship with low-density lipoprotein cholesterol levels rather than the protective effect of high high-density lipoprotein cholesterol levels. Similar results from several modified analyses make low cholesterol level due to severe illness an unlikely explanation for our results. Physicians should be cautious about initiating cholesterol-lowering treatment in men and women above 65 to 70 years of age. Only randomized clinical trials in older people can settle the debate over the efficacy and cost-effectiveness of lipid-lowering interventions for reducing mortality and morbidity in this population.