Currently, a reliable serum biomarker for hepatocellular carcinoma (HCC) has not been established, particularly for early-stage HCC (single tumor < 2 cm). We aimed to investigate diagnostic serum exosomal microRNA (exo-miR) panel for early-stage HCC. Driver oncogenic miR (onco-miR) candidates were selected by integrative analysis of miR expression profiles from three different RNA sequencing datasets of human HCC. Expressions of selected onco-miRs in serum exosome were measured using quantitative real-time PCR. Diagnostic performances of serum exo-miRs for HCC were evaluated in the test cohort (N = 24) and validation cohort (N = 144). Serum exo-miR panels were developed using a logistic regression model, and their diagnostic performance was evaluated. Six promising driver onco-miRs, including miR-25-3p, miR-140-3p, miR-423-3p, miR-1269a, miR-4661-5p, and miR-4746-5p, were identified by integrative analysis of three different RNA sequencing datasets. Among the six candidates, four serum exo-miRs (miR-25-3p, miR-1269a, miR-4661-5p, and miR-4746-5p) showed promising performance in the test cohort with area under the receiving operator curve (AUROC) >0.8. In our validation study, serum exo-miR-4661-5p could diagnose HCC in all stages (AUROC = 0.917), even in early stage (AUROC = 0.923), with a greater accuracy than other candidate serum exo-miRs and serum AFP. The panel composed of exo-miR-4661-5p and exo-miR-4746-5p was identified as the most accurate biomarker for early-stage HCC (AUROC = 0.947, 95% confidence interval = 0.889-0.980, sensitivity = 81.8%, and specificity = 91.7%). In conclusion, exo-miR-4661-5p-based serum panel is a promising diagnostic marker for early-stage HCC.