The role of serum cholesterol in predicting the risk of stroke is unclear and may depend on the subtype of the disease. In 1964 to 1965, 54,385 Swedish men and women participated in a health survey with serum cholesterol and diastolic blood pressure determinations. The Swedish mortality register was used to identify causes of death in this cohort during 20.5 years of follow-up (1964 to 1985). A person-year-based Poisson model was used for multivariate analysis. Relative risk increased with decreasing serum cholesterol level for subarachnoid hemorrhage in men and for cerebral hemorrhage in women but not for subarachnoid hemorrhage in women. For cerebral hemorrhage in men, the risk function was U-shaped. Adjustment for diastolic blood pressure did not significantly change the relation between the risk for any of the different stroke types and the cholesterol level. A low cholesterol level predicts death from intracranial bleeding, but the data suggest that there is differing risk pattern for men and women.