Improving the quality of Cone-Beam Computed Tomography (CBCT) reconstructed images under complicated noise conditions was critical for CBCT systems. In this study, an image reconstruction method of CBCT based on a hybrid Poisson-Gaussian maximum likelihood function was proposed. First, a CBCT image reconstruction model suitable for describing a mixed Poisson-Gaussian noise environment was studied. The model contained a fidelity term based on a mixed Poisson-Gaussian maximum likelihood function and a constraint term based on a three-dimensional total variation regularization method. The fidelity term was used to constrain the reconstruction result to match the observed value as closely as possible in the mixed noise model, where as the constraint term was used for noise removal and preserving the edge and detail information as effectively as possible. The proposed model was further solved using the separable approximation and extended Lagrangian methods. Finally, the effectiveness of the algorithm was verified using both simulation and real data. The results indicate that the proposed method exhibited a maximum improvement of 2.1 dB as compared to other methods evaluated usingsimulated data. In visual comparisons, the proposed method demonstrated the best denoising performance. We can thus conclude that the proposed method is effective forCBCT reconstruction under low dose conditions.