BACKGROUND AND OBJECTIVES: The diagnosis of paroxysmal atrial fibrillation (PAF) and the prediction of its recurrence are sometimes difficult. There have been several recent studies attempting to detect patients at risk for PAF while in sinus rhythm by using the P wave signal-averaged ECG. We undertook to define an appropriate technique of P wave signal-averaged ECG and to estimate the reproducibility of the test. Additionally, we estimated the usefulness of P wave signal-averaged ECG in patients at risk for PAF. SUBJECTS AND METHODS: Forty-five patients with PAF were included in the study undertaken between March 1997 and June 1998. Twelve-lead surface ECG and P wave signal-averaged ECG were performed in the patients. The total P wave duration was measured by the P wave signal-averaged ECG using P wave template and least-square fit filter. The same process was followed in forty sex- and age-matched controls. RESULTS: The measurement of P wave duration with P wave signal-averaged ECG was highly reproducible. The measured P wave duration showed significant prolongation in the patient group at cutoff frequencies of 20 Hz and 30 Hz (123.6+/-15.3 vs. 114.8+/-14.5 msec, p=0.009 at 20 Hz, 120.1+/-17.8 vs. 107.5+/-18.8 msec, p=0.002 at 30 Hz). An abnormal P wave duration defined as over 120 msec in duration by P wave signal-averaged ECG was able to detect PAF with a sensitivity of 60%, specificity of 73%, positive predictive value of 71%, and a negative predictive value of 62%. CONCLUSION: A prolonged P wave duration as measured by P wave signal-averaging technique may be a simple noninvasive marker of risk for the development of atrial fibrillation.