To investigate the relationship between plasma lipids and risk of death from all causes in nondemented elderly. Prospective cohort study. Community-based sample of Medicare recipients, aged 65 years and older, residing in northern Manhattan. Two thousand two hundred seventy-seven nondemented elderly, aged 65 to 98; 672 (29.5%) white/non-Hispanic, 699 (30.7%) black/non-Hispanic, 876 (38.5%) Hispanic, and 30 (1.3%) other. Anthropometric measures: fasting plasma total cholesterol, triglyceride, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and non-HDL-C, body mass index, and apolipoprotein E (APOE) genotype. clinical measures: neuropsychological, neurological, medical, and functional assessments; medical history of diabetes mellitus, heart disease, hypertension, stroke, and treatment with lipid-lowering drugs. Vital status measure: National Death Index date of death. Survival methods were used to examine the relationship between plasma lipids and subsequent mortality in younger and older nondemented elderly, adjusting for potential confounders. Nondemented elderly with levels of total cholesterol, non-HDL-C, and LDL-C in the lowest quartile were approximately twice as likely to die as those in the highest quartile (rate ratio (RR)=1.8, 95% confidence interval (CI)=1.3-2.4). These results did not vary when analyses were adjusted for body mass index, APOE genotype, diabetes mellitus, heart disease, hypertension, stroke, diagnosis of cancer, current smoking status, or demographic variables. The association between lipid levels and risk of death was attenuated when subjects with less than 1 year of follow-up were excluded (RR=1.4, 95% CI=1.0-2.1). The relationship between total cholesterol, non-HDL-C, HDL-C, and triglycerides and risk of death did not differ for older (>or=75) and younger participants (>75), whereas the relationship between LDL-C and risk of death was stronger in younger than older participants (RR=2.4, 95% CI=1.2-4.9 vs RR=1.6, 95% CI=1.02-2.6, respectively). Overall, women had higher mean lipid levels than men and lower mortality risk, but the risk of death was comparable for men and women with comparable low lipid levels. Low cholesterol level is a robust predictor of mortality in the nondemented elderly and may be a surrogate of frailty or subclinical disease. More research is needed to understand these associations.