Elevated plasma total homocysteine level has been associated with cardiovascular disease in many studies, mostly in Europe and North America. Data on persons from other areas and on associations with overall mortality are sparse. To determine the relation of plasma homocysteine level to all-cause and cause-specific mortality. Prospective observational study with 9- to 11-year follow-up. A free-living, multiethnic Jewish population in western Jerusalem, Israel. 1788 residents of Jerusalem (808 men and 980 women) who were at least 50 years of age and were examined between 1985 and 1987 as part of the Kiryat Yovel Community Health Study. Nonfasting plasma homocysteine level was determined in frozen stored samples. Deaths during follow-up were identified by linkage with the national population registry. Plasma homocysteine levels exceeded 14 micromol/L in 28% of men and 20% of women. During the study period, 405 deaths occurred. In multivariate Cox models that controlled for possible confounders, a nonmonotonic increase in mortality hazard ratios was associated with ascending quintile of homocysteine level: 1.0, 1.4, 1.3, 1.5, and 2.0 (P < 0.001 for trend). The relation was similar for cardiovascular and noncardiovascular causes of death (excluding cancer). The association was weaker when deaths that occurred during the first 5 years of follow-up were excluded; corresponding hazard ratios for ascending quintile of homocysteine level were 1.0, 1.0, 1.2, 1.1, and 1.6 (P = 0.063 for trend). Age- and sex-adjusted percentages of deaths "attributable" to elevated plasma homocysteine level (> or = 14 micromol/L) were 12.5% (95% CI, 6.7% to 18.8%) for all deaths, 16.0% (CI, 7.2% to 25.6%) for deaths during the first 5 years of follow-up, and 8.3% (CI, 1.5% to 16.1%) for later deaths. A mildly to moderately elevated nonfasting plasma homocysteine level is a substantial risk marker for death from all causes. The association seems to be stronger during the first 5 years of follow-up.