This article studies the role of dispersion of atrial refractoriness (DAR) in the genesis of atrial fibrillation (AF). A 20-polar Halo catheter or a 40-polar basket catheter was placed in the right atrium and a 10-polar catheter in the coronary sinus in 21 patients with paroxysmal AF. Bipolar electrograms during AF were recorded from 7 to 16 sites in both atria. As control, electrograms during AF induced by extra-stimulation or burst pacing were also recorded from 4 to 14 sites in both atria in 12 patients with supraventricular tachycardias but without history of AF. The local atrial fibrillation intervals (AFI) during a period of 10 s or 20 s were measured and the mean, median and the 5th, 10th and 15th percentile AFIs at each site were calculated as estimates of the local effective refractory period (AERP). The maximum dispersion and variance of the estimated AERP among the 7-16/4-14 sites were used as measures of the DAR. The maximum dispersion and variance of the 5th and 10th percentile AFIs were significantly greater in the AF group than those in the control group, which were mainly due to the shortening of the minimum 5th and 10th percentile AFIs. No significant differences in dispersion and variance of the mean and median AFIs were shown between the 2 groups. The dispersion and variance of atrial refractoriness during AF estimated from the measurement of short AFIs were significantly greater in patients with paroxysmal AF than in those without clinical AF. The increased dispersion of refractoriness in patients with AF was mainly due to the shortening of the minimum AFIs. These findings suggest the involvement of an increased dispersion of atrial refractoriness in the genesis of paroxysmal AF.