The clinical outcome of hepatocellular carcinoma (HCC) treatment remains unsatisfactory, contributing to the high mortality of HCC worldwide. Circulating miRNAs have the potential to be a predictor of therapy response. Microarray profiling was performed in serum samples of 20 HCC patients before treatment. Circulating miRNAs associated with treatment response were validated in 86 serum HCC samples using the qRT-PCR system. Patients were treated either with curative treatments (resection or radiofrequency) or trans-arterial chemoembolization (TACE), and grouped according to therapy response in complete responders (CR) and partial responders or progressive disease (PRPD), following mRECIST criteria. Four miRNA candidates from the discovery phase (miR-4443, miR-4454, miR-4492, and miR-4530) were validated. Before therapy, miR-4454 and miR-4530 were up-regulated in CR to curative treatments (2.83 fold, p = 0.02 and 2.33 fold, p = 0.008, respectively) and were able to differentiate CR from PRPD (area under the curve (AUC) = 0.74, sens/spec 79/63% and AUC = 0.77, sens/spec 72/73%). On the contrary, miR-4443 was 1.95 times down-regulated in CR (p = 0.05) with an AUC of 0.72 (sens = 70%, spec = 60%) in distinguishing CR vs. PRPD). The combination of the three miRNAs was able to predict the response to curative treatment with an AUC of 0.84 (sens = 72%, spec = 75%). The higher levels of miR-4454 and miR-4530 in were associated to longer overall survival (HR = 2.79, p = 0.029 and HR = 2.97, p = 0.011, respectively). Before TACE, miR-4492 was significantly up-regulated in CR patients (FC = 2.67, p = 0.01) and able to differentiate CR from PRPD (AUC = 0.84, sens/spec 84.6/71%). We demonstrated that different miRNAs predictors can be used as potential prognostic circulating biomarkers according to the selected treatment for HCC.