The mode of onset of atrial fibrillation has been recorded 18 times in 14 patients during continuous monitoring of an electrocardiogram. An atrial premature discharge preceded the onset of atrial fibrillation in all 14 episodes of relapse from sinus rhythm. In 10 of these the premature P wave was followed immediately by atrial fibrillation. In 4 the premature P wave appeared to initiate a brief run of atrial tachycardia (less than 6 beats) which accelerated to atrial fibrillation. In 4 observations, atrial fibrillation developed from established atrial flutter. The atrial rate gradually accelerated from approximately 300 per minute, and irregular, rapid atrial response characteristic of atrial fibrillation appeared. In 1 patient this event occurred during carotid sinus massage. The relative prematurity of ectopic atrial beats was evaluated by calculating the coupling index, defined as the ratio of the coupling interval (P-P) to the preceding cycle length. The mean coupling index of the 14 atrial premature contractions which initiated atrial fibrillation was .48 ± .03 (S.E.M.). The mean coupling index for atrial premature contractions which were not followed by atrial fibrillation was .65 ± .02 in the patients who subsequently developed atrial fibrillation, and .68 ± .01 in a similar group of patients with ectopic atrial activity who did not develop fibrillation. The difference between the mean coupling indices of those P waves which initiated atrial fibrillation and those which did not is highly significant (p < .001). It is concluded that a spontaneously occurring atrial premature impulse may initiate atrial fibrillation. In some instances the fibrillation is preceded by an accelerating atrial tachycardia. The propensity of a premature impulse to initiate fibrillation is related to its relative prematurity, expressed as the coupling index. When the coupling index is less than .50, the change of atrial fibrillation is high; when it is greater than .60, the chance of atrial fibrillation following that particular premature P wave is small. The observation that atrial fibrillation may follow a properly timed ectopic discharge is consistent with the experimental delineation of an atrial vulnerable period. Possible relationships between focal ectopic activity, atrial tachycardias, and atrial fibrillation are discussed.