Background: The detection of underlying paroxysmal atrial fibrillation (AF) in patients with cryptogenic stroke (CS) can be challenging, and there is great interest in finding predictors of its hidden presence. The recent development of sophisticated software has enhanced the agnostic and prognostic performance of the 12-lead electrocardiogram (ECG). Our aim was to assess the additional role of a computer-assisted ECG analysis in identifying predictors ofAF in patients with CS. Methods: Sixty-seven patients with ischemic stroke or high-risk transient ischemic attack of unknown etiology were prospectively studied. Their 12-lead digitized ECG was analyzed with dedicated software, quantifying 468 morphological variables. The main clinical, biochemical, and echocardiographic variables were also collected. At discharge, patients were monitored with a wearable Holter for days, and the primary outcome was the detection of AF. Results: The median age was 80 (interquartile range (IQR): 73-84) and AF was detected in 21 patients (31.3%). After preselecting significant ECG variables from the univariate analysis, a multivariate regression including other significant clinical, biochemical and echo-cardiographic predictors of AF was performed. Among the automatically analyzed ECG parameters, the amplitude of the R wave in V1 (V1_ramp) was significantly associated with the outcome. The best model to predict AF was composed of age, N-terminal B-type natriuretic peptide (NT-proBNP), left atrial reservoir strain (LASr) and V1_ ramp. This model showed good discrimination capacity (corrected Somer's Dxy 0.941) and performed better than the same model without the ECG variable (Somer's Dxy: 0.827, Brier's B: 0.119,AUC: 0.896). : 0.907, Brier's B: 0.079, area under the curve (AUC): Conclusions: The addition of computer-assisted ECG analysis can help stratify the risk ofAF in the challenging clinical setting of CS.