Purpose: Hepatocellular cancer (HCC) is the sixth most prevalent cancer and the third leading cause of cancer-related death worldwide. Cellular immunotherapy against glypican 3 (GPC3) has recently been used in the treatment of HCC, following the success of chimeric antigen receptor (CAR)-T therapy in treatment of B cell malignancy. However, CART cells are not "off-the-shelf" and always cause cytokine release syndrome, which can be eliminated by using natural killer (NK) cells as effector cells. Since a costimulatory signal is necessary for the activation, persistence, or cytotoxicity of CART cells, we speculated that the costimulatory signal is also required for CAR-NK cells in HCC treatment. Methods: Five anti-GPC3 CAR plasmids containing different costimulatory domains were constructed. They included Z (only the CD3ζ domain, no costimulatory domain), CD28.Z (T-cell costimulatory domain CD28), DNAM1/2B4.Z (NK-cell-associated costimulatory domain DNAM1 or 2B4), and DNAM1.2B4.Z (both NK-cell-associated costimulatory domains). Respective CAR-NK-92 cells were generated. The MTT viability assay was performed to evaluate the effect of the different costimulatory domains on CAR-NK-cell proliferation. The effect on persistence was analyzed using an apoptosis assay and flow cytometry. Special cytotoxicity against normal hepatocellular cells and GPC3 + malignant cells was investigated in vitro. The concentration of cytokines (TNF-α and IFN-γ) released by CAR-NK-92 cells was also measured by ELISA. Results: NK-cell-associated costimulatory signal was necessary for CAR-NK-92 cells. CAR-NK-92 cells with DNAM1 and/or 2B4 expanded more quickly and persisted with a lower apoptotic ratio, compared to the presence of CD28 or no costimulatory signal. All CAR-NK-92 cells showed special cellular cytotoxicity in vitro. CAR-NK-92 cells with NK-cell-associated costimulatory domains exhibited higher cytotoxic ability compared with those without any costimulatory domain or with T-cell costimulatory domain. CAR-NK-92 cells with both DNAM1 and 2B4 displayed the highest cytotoxicity. The cytokine release assay results were consistent with those of the cytotoxicity assay. Conclusion: We provided the first evidence supporting a strategy using DNAM1 and 2B4 costimulatory domains to generate anti-GPC3 CAR-NK-92 cells, which exhibits enhanced cytotoxicity against hepatocellular cancer cells in vitro.