The BLISS and Berten’s prognostic scores are the only models developed using best-practice methodology to predict COPD exacerbations. However, they require testing in settings outside of primary care to confirm transportability. To validate the BLISS and Berten’s scores in the ECLIPSE COPD cohort. We validated both scores for predicting 2-year exacerbation with 1-year and 3-year exacerbation as secondary endpoints. Area-under-the-curve (AUC), calibration slope, and intercept/calibration-in-the-large (CITL) determined discrimination, under-or-over-prediction of risk in certain individuals or subgroups, and systematic under-or-over-prediction of risk, respectively. Of 1,749 included patients (64.8% male, mean age = 63 years, mean FEV1 = 49% of predicted), 1214 (69%) had >1 exacerbation within 2-years. The BLISS score had good discrimination (AUC= 0.73; 95% CI: 0.71-0.76) and performance remained stable at 1- and 3-years (AUC= 0.72 and 0.74, respectively). Calibration slopes improved across time periods (year 1=0.87, year 2=0.92, and year 3=0.97). Berten’s score was less accurate at 2-years (AUC= 0.70; 95% CI: 0.67 to 0.73) and at 1- and 3-years (AUC= 0.70 and 0.71, respectively). Calibration slopes for Berten9s score also improved with increasing time periods (year 1=0.81, year 2=0.84, and year 3=0.97). Berten’s score showed systematic under-prediction of risk (2-year CITL = 1.36) but the BLISS score showed worse systematic under-prediction (2-year CITL = 2.30). Both scores showed accurate discrimination. However, they exhibited systematic under-prediction of risk, especially the BLISS score. Therefore, both scores may need their intercepts updated to be valid in more severe COPD populations.