BACKGROUND: The risk for atrial fibrillation-associated stroke increases at low anticoagulation intensities. However, higher intensities increase hemorrhage risk. Optimal use of warfarin for atrial fibrillation requires precise information on the risk for intracranial hemorrhage as a function of patient age and anticoagulation intensity. OBJECTIVE: To examine the relationship of age, anticoagulation intensity, and risk for intracranial hemorrhage. DESIGN: Case-control study. SETTING: Academic medical center. PATIENTS: 170 case-patients who developed intracranial hemorrhage during warfarin therapy and 1020 matched controls who did not; both case-patients and controls were taking warfarin for atrial fibrillation. MEASUREMENTS: The authors performed multivariable conditional logistic regression to determine the odds of intracranial hemorrhage with regard to age and international normalized ratio (INR), controlling for comorbid conditions and aspirin use. RESULTS: Case-patients were older than controls (median age, 78 years vs. 75 years; P < 0.001) and had higher median INRs (2.7 vs. 2.3; P < 0.001). The risk for intracranial hemorrhage increased at 85 years of age or older (adjusted odds ratio, 2.5 [95% CI, 1.3 to 4.7]; referent age, 70 to 74 years) and at an INR range of 3.5 to 3.9 (adjusted odds ratio, 4.6 [CI, 2.3 to 9.4]; referent INR, 2.0 to 3.0). The risk for intracranial hemorrhage at INRs less than 2.0 did not differ statistically from the risk at INRs of 2.0 to 3.0 (adjusted odds ratio, 1.3 [CI, 0.8 to 2.2]). LIMITATIONS: Although duration of anticoagulation has been associated with hemorrhage in other studies, the current study could not control for this potential confounder. CONCLUSIONS: The risk for intracranial hemorrhage increases at age 85 years. International normalized ratios less than 2.0 were not associated with lower risk for intracranial hemorrhage compared with INRs between 2.0 and 3.0. Therefore, anticoagulation management should focus on maintaining INRs in the 2.0 to 3.0 range, even in elderly patients with atrial fibrillation, rather than targeting INRs less than 2.0. Similarly, INRs of 3.5 or greater should be avoided.