Unique change of circulating microRNAs (miRNAs) was recognized to occur in early oncogenesis, which conferred it the potential as biomarkers for early detection of cancer. However, its diagnostic capability for hepatocellular carcinoma (HCC) has not been fully understood. In this study, microarray analysis was applied to screen the initial candidate miRNA from both the supernatants of anoikis-resistant cellular models and the sera samples of HCC patients. The selected differentially expressed miRNAs were further verified in 115 HCC patients and 40 health controls by qRT-PCR. Among these, 4 miRNAs (miR-16-2-3p, 92a-3p, 107, and 3126-5p) were significantly changed in HCC patients compared with controls. Logistic regression analysis identified a 3-miRNA panel (miR-92-3p, miR-107, and miR-3126-5p) as valuable diagnostic marker for HCC, especially for early stage patients (AUC = 0.975) and for low-level AFP HCC patients (AUC = 0.971). In addition, the combination of 3-miRNA panel and AFP was even more effective for discriminating the early stage HCC patients (AUC = 0.988) and low-level AFP HCC patients (AUC = 0.989) from control. In conclusion, diagnostic efficacy of the combination of 3-miRNA panel and AFP was powerful for HCC diagnosis, especially in early tumor screening and low-level AFP patients.