Alpha-fetoprotein (AFP) is the most widely used biomarker for hepatocellular carcinoma (HCC) surveillance, which is criticized as neither sensitive nor specific in active hepatitis and liver cirrhosis. The aim of this study was to determine the performance of AFP as a tumor marker for HCC in entecavir-treated patients with chronic hepatitis B (CHB). This was a retrospective-prospective cohort study of 1,531 entecavir-treated patients under regular HCC surveillance with AFP and ultrasonography. Mean age was 52 ± 12 years; 1,099 (72%) patients were male and 332 (21.7%) had clinical evidence of cirrhosis. At a mean follow-up of 51 ± 13 months, 57 (2.9%) patients developed HCC (median size: 3.3 cm). AFP fluctuated with alanine aminotransferase (ALT) and peaked at the time of starting entecavir, then gradually decreased after. AFP started to increase 6 months before the diagnosis of HCC. The receiver operator characteristic curve (AUROC) of AFP was highest at the time of HCC diagnosis (0.85; 95% confidence interval [CI]: 0.73-0.98) and remained satisfactory at 3 (0.82; 95% CI: 0.73-0.91) and 6 months (0.79; 95% CI: 0.69-0.89) before the diagnosis. Using the conventional AFP cut-off (20 μg/L) at month 0, the sensitivity and specificity to diagnose HCC were 38.6% and 98.9%, respectively. Adopting the lower cut-off value (6 μg/L) of AFP level at month 0, sensitivity was increased to 80.7%, whereas specificity was decreased to 80.4%. Conclusion: On-treatment AFP is a specific tumor marker for HCC in CHB patients receiving entecavir therapy. Adopting a lower cut-off value of AFP level at 6 μg/L would significantly increase the sensitivity for HCC detection. (Hepatology 2014;59:986–995)