Previous work has suggested that a comparison of electrograms from two or more sites may best differentiate fibrillatory from nonfibrillatory rhythms. The coherence spectrum is a measure by which two signals may be compared quantitatively in the frequency domain. In the present study, the coherence spectrum was used to quantify the relation between spectral components of electrograms from two sites in either the atrium or ventricle during both fibrillatory and nonfibrillatory rhythms. Bipolar recordings of 35 rhythms from 20 patients were analyzed for coherence in the 1-59 Hz band. The 17 nonfibrillatory rhythms were sinus rhythm (six), paroxysmal supraventricular tachycardia (two), atrial flutter (four), and monomorphic ventricular tachycardia (five). The 18 fibrillatory rhythms were atrial fibrillation (12) and ventricular fibrillation (six). Nonfibrillatory rhythms exhibited moderate-to-high levels of coherence throughout the 1-59 Hz band, with peaks concentrated at the rhythm's fundamental frequency and its harmonics. Fibrillatory rhythms exhibited little coherence throughout the 1-59 Hz band, and harmonics were not evident. The mean magnitude-squared coherence (scale of 0 to 1) for the 1-59 Hz band ranged from 0.22 to 0.86 (mean +/- SD, 0.52 +/- 0.19) for nonfibrillatory rhythms and from 0.042 to 0.12 (0.067 +/- 0.021) for fibrillatory rhythms. Separation of fibrillatory and nonfibrillatory rhythms was possible whether signals were recorded by floating or fixed-electrode configurations. These findings indicate that comparison of two electrograms with magnitude-squared coherence measurements differentiates fibrillatory from nonfibrillatory rhythms. A recognition algorithm based on coherence spectra may provide a major variations in lead configuration.(ABSTRACT TRUNCATED AT 250 WORDS)