Objectives: Predicting hepatocellular carcinoma (HCC) in patients with chronic hepatitis B who received long-term therapy with potent nucleos(t)ide analogs is of utmost importance to refine the strategy for HCC surveillance. Methods: We conducted a multicenter retrospective cohort study to validate the CAGE-B and SAGE-B scores, HCC prediction models developed for Caucasian patients receiving entecavir (ETV) or tenofovir (TFV) for >5 years. Consecutive patients who started ETV or TFV at two hospitals in Korea from January 2009 to December 2015 were identified. The prediction scores were calculated, and model performance was assessed using receiver operating characteristics (ROC) curves. Results: Among 1557 patients included, 57 (3.7%) patients had HCC during a median follow-up of 93 (95% confidence interval, 73–119) months. In the entire cohort, CAGE-B predicted HCC with an area under the ROC curve of 0.78 (95% CI, 0.72–0.84). Models that have “liver cirrhosis” in the calculation, such as AASL (0.79 (0.72–0.85)), CU-HCC (0.77 (0.72–0.82)), and GAG-HCC (0.79 (0.74–0.85)), showed accuracy similar to that of CAGE-B (p > 0.05); however, models without “liver cirrhosis”, including SAGE-B (0.71 (0.65–0.78)), showed a lower predictive ability than CAGE-B. CAGE-B performed well in subgroups of patients treated without treatment modification (0.81 (0.73–0.88)) and of male sex (0.79 (0.71–0.86)). Conclusions: This study validated the clinical usefulness of the CAGE-B score in a large number of Asian patients treated with long-term ETV or TFV. The results could provide the basis for the reappraisal of HCC surveillance strategies and encourage future prospective validation studies with liver stiffness measurements.