OBJECTIVE: To investigate risk factors, such as heavy alcohol consumption, that might explain any increased risk of haemorrhagic stroke associated with low blood cholesterol. DESIGN: Prospective cohort study. SETTING: Korea. PARTICIPANTS: 787,442 civil servants (661,700 men, 125,742 women) aged 30-64. MAIN OUTCOME MEASURES: Cardiovascular risk factors were assessed at biennial health check. Data on morbidity and mortality were ascertained from 1990 to 2001 using hospital admissions and mortality surveillance systems. RESULTS: 6328 cases of ischaemic stroke (6021 men, 307 women), 3947 cases of haemorrhagic stroke (3748 men, 199 women), 3170 cases of undefined stroke (2902 men, 268 women), and 4417 cases of myocardial infarction (4305 men, 112 women) occurred. Ischaemic stroke and myocardial infarction were strongly and positively associated with blood cholesterol (hazard ratio per 1 mmol/l cholesterol 1.20 (95% confidence 1.16 to 1.24) and 1.48 (1.43 to 1.53), respectively). Haemorrhagic stroke showed an inverse association in fully adjusted models (0.91, 0.87 to 0.95). This inverse association was confined to participants with hypertension. When stratified by concentration of gamma glutamyl transferase (GGT), an indicator of alcohol consumption, the association was not seen in participants with low concentrations of GGT, and it was independent of hypertension in those with high concentrations of GGT (> 80 U/l).