Cardioembolic cerebral infarction is a subtype of stroke with a high mortality. The purpose of this study was to determine predictors of in-hospital mortality in 231 consecutive patients with cardioembolic stroke by means of a multivariate analysis. Three predictive models were constructed. A first model was based on demographic, anamnestic and clinical variables collected at the bedside examination (total 8 variables). A second model was based on clinical and neuroimaging variables (total 10 variables). A third model was based on the aforementioned clinical and neuroimaging variables and the presence of early recurrent embolism (total 11 variables). Deteriorated level of consciousness, limb weakness, presence of congestive heart failure, male gender, and age appeared to be independent prognostic factors of in-hospital mortality in the predictive model based on clinical variables and in the predictive model based on clinical and neuroimaging variables. In addition to these variables, early recurrent embolization was selected in the third predictive model. In the first two models, setting a cut-off point of 0.50 for predicting vital status at hospital discharge resulted in a sensitivity of 60%, a specificity of 89% and a total correct classification of 81%. The corresponding values of the third model were 62, 89 and 81%, respectively. These data may help clinicians to establish an early prognosis of this stroke subtype more accurately as well as to allocate patients with cardioembolic stroke in clinical trials correctly.