Background: The use of a severity score to help orientation decisions could improve the efficiency of care for acute exacerbations of COPD (AECOPD). We previously developed a score ('2008 score', based on age, dyspnea grade at steady state and number of clinical signs of severity) predicting in-hospital mortality in patients with AECOPD visiting emergency departments (EDs). External validity of this score remained to be assessed. Objectives: To test the predictive properties of the '2008 score' in a population of patients hospitalized in medical respiratory wards for AECOPD, and determine whether a new score specifically derived from this population would differ from the previous score in terms of components or predictive performance. Methods: Data from a cohort study in 1824 patients hospitalized in a medical ward for an AECOPD were analyzed. Patients were categorized using the 2008 score and its predictive characteristics for in-hospital mortality rates were assessed. A new score was developed using multivariate logistic regression modeling in a randomly selected derivation population sample followed by testing in the remaining population (validation sample). Robustness of results was assessed by case-by-case validation. Results: The 2008 score was characterized by a c-statistic at 0.77, a sensitivity of 69% and a specificity of 76% for prediction of in-hospital mortality. The new score comprised the same variables plus major cardiac comorbidities and was characterized by a c-statistic of 0.78, a sensitivity of 77% and specificity of 66%. Conclusions: A score using simple clinical variables has robust properties for predicting the risk of in-hospital death in patients hospitalized for AECOPD. Adding cardiac comorbidities to the original score increased its sensitivity while decreasing its specificity.