Background: Tracking down atrial fibrillation (AF) in the stroke unit is a relevant challenge for the prevention of recurrent AF-related stroke. The optimal terms of use of continuous ECG monitoring (CEM) are unknown. We compared 24-hour routine Holter ECG with two different CEM analysis strategies for AF detection. Methods: We prospectively enrolled consecutive ischemic stroke patients. All AF-naïve patients received CEM during hospitalization. Two methods for reading CEM data were compared: manual analysis using the Holter function (hCEM) and semiautomated analysis using software (aCEM). The McNemar test was used to compare AF detection rates. Results: Of the 362 patients included, 58 (16.0%) were non-AF-naïve patients and 304 were AF-naïve patients. AF-Naïve patients underwent CEM with a median duration of 5.3 days (3.4-9.7). We detected 22 new AF cases (7.2%) with first-24-hour hCEM, 31 (10.2%) with aCEM, and 42 (13.8%) with hCEM. hCEM and aCEM both significantly increased the AF detection rate compared to first-24-hour hCEM. hCEM detected more new AF cases than aCEM (+3.6%, p = 0.003). Conclusions: In stroke patients, early and prolonged aCEM and hCEM both increase the AF detection rate compared to first-24-hour hCEM. hCEM gives the best AF detection rate. We suggest that in aCEM, detection based only on the ventricular rhythm analysis explains its lower specificity and sensitivity.