Despite advances in imaging techniques and detailed examinations to determine the etiology of a stroke, the cause still remains undetermined in about one fourth of all ischemic strokes. The aim of this prospective study was to determine whether perfusion magnetic resonance imaging (MRI) can differentiate cardioembolic stroke from large artery atherosclerosis (LAA). We recruited 17 cardioembolic stroke and 22 LAA stroke patients, who were classified according to the Trial of Org 10172 in Acute Stroke Treatment and underwent perfusion MRI within 24 hours after the onset of stroke. The patients with cardioembolic stroke had more severe initial stroke severity and larger volumes of initial and final infarct compared to those with LAA stroke. Receiver operating characteristic curve analysis showed that the ratio of time to maximum of the residual curve (Tmax) volume for a 2-, 3-, 4- or 5-s lag over Tmax volume for a 8s lag all had excellent area under the curve values (> 0.9) to predict cardioembolic stroke. After adjusting for initial National Institute of Health Stroke Scale scores, a threshold of 3.73 for (Tmax > 4s volume)/(Tmax > 8s volume) had the highest odds ratio to predict cardioembolic stroke (p=0.012; odds ratio: 58.5; 95% confident interval: 2.5-1391.1), with 87.5% sensitivity and 94.4% specificity. In conclusion, perfusion MRI could be a reliable tool to identify cardioembolic stroke with its lower collateral. This is important as it could be used to reveal the exact mechanism and provide supportive evidence to classify a stroke.