Background: Detection of paroxysmal atrial fibrillation (PAF) in acute ischemic stroke patients poses diagnostic challenge. The aim of this study was to predict the presence of PAF by means of 12-lead ECG in patients with acute ischemic stroke. Our hypothesis was that P-wave dispersion (Pd) might be a useful marker in predicting PAF in patients with acute ischemic stroke. Methods: 12-lead resting ECGs, 24-hour Holter recordings and echocardiograms of 400 patients were analyzed retrospectively. PAF was detected in 40 patients on 24-hour Holter monitoring. Forty out of 360 age and gender matched patients without PAF were randomly chosen and assigned as the control group. Demographics, P-wave characteristics and echocardiographic findings of the patients with and without PAF were compared. Results: Maximum P-wave duration (p=0.002), Pd (p<0.001) and left atrium diameter (p=0.04) were significantly higher in patients with PAF when compared to patients without PAF. However, in binary logistic regression analysis Pd was the only independent predictor of PAF. The cut-off value of Pd for the detection of PAF was 57.5 milliseconds (msc). Area under the curve was 0.80 (p<0.001). On a single 12-lead ECG, a value higher than 57.5 msc predicted the presence of PAF with a sensitivity of 80% and a specificity of 73%. Conclusion: Pd on a single 12-lead ECG obtained within 24 hours of an acute ischemic stroke might help to predict PAF and reduce the risk of recurrent strokes.