OBJECTIVES: Studies assessing the effectiveness of new Chronic Obstructive Pulmonary Disease (COPD) treatments commonly use disease-specific health-related quality-of-life (HR-QoL) instruments such as the Clinical COPD Questionnaire (CCQ). However, for the economic evaluations, utility data is necessary. This study aims to develop a model to predict mean utility values for a group of COPD patients using CCQ data. METHODS: We combined the data from two trials (RECODE and GO-AHEAD) including over 5000 observations with a broad range of disease severity. Data was randomly split into an estimation and validation sample. The overlap between the CCQ and EQ-5D was assessed using principal component and correlation matrix analysis. Different types of models were created with increasing complexity. We analysed the effect of using different observations of the same patients as unique observations and performed a sensitivity analysis using different EQ-5D value sets to estimate EQ-5D utilities. The external validity was tested with the dataset of the MARCH trial.  RESULTS: The principal component analyses showed a poor correlation of the dimension pain/discomfort with any of the CCQ items and the CCQ items cough and produce phlegm did not load onto any EQ-5D dimension. The model using ordinary least square with the individual CCQ items as dummy variables, controlled for sex, resulted in the best predicted performance. The mean absolute error was on average 0.15 but was considerably high (>0.34) if the observed EQ-5D value was below 0.5. Sensitivity analysis revealed that different EQ-5D value sets resulted in different algorithms but similar predicting ability. CONCLUSIONS: Our study showed that there are conceptual differences between the CCQ and EQ-5D and mapping should be considered as second-best option compared to directly collected EQ-5D data. Furthermore, the mapping performance seems to depend on the severity of the study population.