This study tests the hypothesis that atrial fibrillation (AFib) can be discriminated from regular atrial rhythms by a measure of the variation in local activation direction. Human endocardial atrial recordings of AFib, sinus rhythm, atrial flutter, and supraventricular tachycardia were collected using a catheter with orthogonally placed electrodes, and the direction of each activation was calculated using methods previously described by our laboratory. Each recording was divided into segments containing 100 activations, and the spatial precision for each segment was calculated in three dimensions, as well as in each of the three two-dimensional (2-D) planes. The three-dimensional (3-D) spatial precision for 1161 segments of AFib in 11 recordings ranged from 0.09-0.85 (mean = 0.45), whereas the spatial precision for 138 segments of regular rhythms in 28 recordings was > or = 0.91 in all but four instances. The 2-D spatial precision values overlapped for all rhythms. The results indicate that 3-D spatial precision of local activation direction is a useful discriminator of AFib.