Five randomized trials of warfarin stroke prophylaxis in atrial fibrillation have undergone meta-analyses by the Atrial Fibrillation Investigators (AFI) and by the British Columbia Office of Health Technology Assessment (BCOHTA), with differing conclusions. The AFI, using the original data, applied a consistent definition of 'major' bleeding (intracranial, hospitalization or transfusion of at least 2 U of blood) and found an excess of six major bleeding events. The BCOHTA used the definitions used in the studies, including "any medical intervention", and counted an excess of 21 'major' bleeding events. They then compared these with only the most severe one-third of the strokes. The BCOHTA were concerned that lack of blinding may have influenced the diagnosis of mild stroke, but the data do not suggest diagnostic bias. The risk reduction in the BCOHTA analysis of the most severe one-third of strokes was almost identical to that in the remaining strokes. The value of treatment is best assessed by comparing good with bad events of similar impact, and eliminating strokes from analysis does not eliminate them from patients. The BCOHTA analysis confirms the risk reduction demonstrated by the AFI.