Recent randomized trials have consistently demonstrated the marked efficacy of warfarin in reducing the risk of stroke caused by nonrheumatic atrial fibrillation. These trials have provided conflicting evidence on the effect of aspirin. We report the aspirin analysis from the BAATAF study, a trial in which control patients could choose to take aspirin. There we two strokes in 446 person-years with warfarin (annual rate of 0.45%); eight strokes in 206 person-years with aspirin, most at 325 mg per day (annual rate of 3.9%); and five strokes in 271 person-years among patients taking neither aspirin nor warfarin (annual rate of 1.8%). Simultaneously controlling for the other significant determinants of stroke in the BAATAF study (age, mitral annular calcification, and clinical heart disease), the relative rates (95% confidence interval) of stroke were: (1) warfarin/aspirin = 0.135 (0.029 to 0.64); (2) aspirin/(no aspirin and no warfarin) = 1.95 (0.64 to 5.97); and (3) warfarin/(no aspirin and no warfarin) = 0.263 (0.051 to 1.36). Our "treatment received" analysis argues that warfarin is strikingly more effective than aspirin in preventing stroke in nonrheumatic atrial fibrillation.