BACKGROUND: Total homocysteine levels are associated with arteriosclerotic outcomes. OBJECTIVE: To determine whether total homocysteine levels predict incident stroke in elderly persons. DESIGN: Prospective population-based cohort study with 9.9 years of follow-up. SETTING: Framingham, Massachusetts. PATIENTS: 1947 Framingham Study participants (1158 women and 789 men; mean age +/- SD, 70 +/- 7 years). MEASUREMENTS: Baseline total homocysteine levels and 9.9-year stroke incidence. RESULTS: The quartiles of nonfasting total homocysteine levels were as follows: quartile 1, 4.13 to 9.25 micromol/L; quartile 2, 9.26 to 11.43 micromol/L; quartile 3, 11.44 to 14.23 micromol/L; quartile 4, 14.24 to 219.84 micromol/L. During follow-up, 165 incident strokes occurred. In proportional hazards models adjusted for age, sex, systolic blood pressure, diabetes, smoking, and history of atrial fibrillation and coronary heart disease, relative risk (RR) estimates comparing quartile 1 with the other three quartiles were as follows: quartile 2 compared with quartile 1--RR, 1.32 (95% CI, 0.81 to 2.14); quartile 3 compared with quartile 1--RR, 1.44 (CI, 0.89 to 2.34); quartile 4 compared with quartile 1--RR, 1.82 (CI, 1.14 to 2.91). The linear trend across the quartiles was significant (P < 0.001). CONCLUSION: Nonfasting total homocysteine levels are an independent risk factor for incident stroke in elderly persons.