Aims To evaluate the clinical inference of serum alpha-fetoprotein (AFP) response in hepatocellular carcinoma (HCC) patients undergoing percutaneous radiofrequency ablation (RFA). Materials and methods Three hundred and thirteen previously untreated HCC patients were enrolled in the study. The optimal AFP response was defined as >20% decrease from baseline after 1 month of RFA for those with a baseline AFP level of ≥100 ng/ml. The impact of AFP response on prognosis was analysed and prognostic factors were assessed. Results After a median follow-up of 26.7 ± 19.1 months, 49 patients died and 264 patients were alive. The cumulative 5 year survival rates were 75.3 and 57.4% in patients with an initial AFP of <100 ng/ml and ≥100 ng/ml, respectively ( p  = 0.003). In the 58 patients with a baseline AFP of ≥100 ng/ml and initial completed tumour necrosis after RFA, the cumulative 5 year survival rates were 62.4 and 25.7% in optimal and non-optimal AFP responders, respectively ( p  = 0.001). By multivariate analysis, the prothrombin time international normalized ratio >1.1 ( p  = 0.009), non-optimal AFP response ( p  = 0.023), and creatinine >1.5 mg/dl ( p  = 0.021) were independent risk factors predictive of poor overall survival. Besides, the cumulative 5 year recurrence rates were 83.4 and 100% in optimal and non-optimal AFP responders, respectively ( p  < 0.001). Multivariate analysis demonstrated platelet count ≤105 /mm3 ( p  = 0.048), tumour size >2 cm ( p  = 0.027), and non-optimal AFP response ( p  < 0.001) were independent risk factors associated with tumour recurrence after RFA. Conclusions Serum AFP response may be a useful marker for predicting prognosis in HCC patients undergoing RFA.