We hypothesized that the variance of P wave duration (P variance) in the 12-lead ECG could reflect the spatial dispersion of P wave duration due to inhomogeneous and delayed propagation of sinus impulses in the atria, and moreover could present better reproducibility than maximum P wave duration and P wave dispersion that have already been used for the prediction of idiopathic paroxysmal AF. We also tested a semiautomated PC-based method to improve the accuracy and reproducibility of P wave measurements. A 12-lead ECG was obtained from 60 patients with idiopathic paroxysmal AF and from 50 healthy controls. All ECGs were analyzed manually using magnifying lens and calipers, while 20 randomly selected ones were scanned and analyzed on screen using common commercial software. P maximum, P dispersion, and P variance were all significantly higher in patients with paroxysmal AF than in controls. A P maximum value of 110 ms, a P dispersion value of 40 ms, and a P variance value of 120 ms2 separated patients from controls with a sensitivity of 88%, 83%, and 80%, respectively and a specificity of 75%, 85%, and 74%, respectively. The reproducibility of P variance was higher compared to P dispersion and P maximum. Finally, the PC-based method significantly increased accuracy and reproducibility of P wave measurements. Thus, the variance of P wave duration could be a useful ECG marker for the prediction of paroxysmal idiopathic AF and the use of PC-based methods may enhance the accurate measuring of P wave duration on the ECG.