This study assesses a technique for quantifying the frequency spectrum of atrial fibrillation (AF) using the surface electrocardiogram. Electrocardiograhic recordings were obtained in 61 patients during AF. After bandpass filtering, the QRST complexes were subtracted using a template-matching algorithm. The resulting fibrillatory baseline signal was subjected to Fourier transformation and displayed as a frequency power spectrum. These frequency spectra were compared to direct measurements from the right atrium and coronary sinus in 35 patients undergoing electrophysiologic study. The clinical use of this technique was explored by correlating fibrillatory frequency with the behavior of the arrhythmia in 26 patients referred for cardioversion. The electrocardiographic frequency spectrum during AF was characterized by a single peak that varied widely between patients (range 228 to 480 beats/min). There was a strong correlation between electrocardiographic peak frequency and that measured in the right atrium and coronary sinus (r = 0.79 to 0.98, p <0.0001). Episodes of AF that terminated in < 5 minutes had a lower frequency than those that persisted > 5 minutes (324 +/- 36 vs 402 +/- 78 beats/min, p = 0.001). Chronic AF (< 3 months in duration) had a lower frequency than chronic AF (present > 3 months) (336 +/- 48 vs 408 +/- 60 beats/ min, p = 0.012). Fibrillation frequency was an accurate predictor of conversion with ibutilide. Success rate was 100% in patients with peak frequency < 360 beats/min versus 29% in patients with frequencies > or = 360 beats/min (p = 0.003). Automatic analysis of the frequency content of the fibrillatory baseline on the surface electrocardiogram accurately reflects the average rate of AF. This measurement correlates with the clinical pattern of the arrhythmia and predicts the response to administration of ibutilide.