Fluorescence molecular tomography is a promising imaging modality and has developed fast in the past years. However, challenges remain in its reconstruction, especially for the reconstruction accuracy and computational efficiency. Generally, an analytical method is computationally efficient while a numerical method has advantages in the reconstruction accuracy for arbitrary geometries. To achieve high reconstruction accuracy at low computational cost, a hybrid method that combines an analytical method based on Kirchhoff approximation (KA) and a numerical method based on finite element method (FEM) is proposed. This method is tested with numerical simulations and phantom experiments. Results of numerical simulations indicate that with the hybrid method, the reconstruction accuracy is improved while the computational time decreases by 40-70% compared with the standalone KA method and FEM. Phantom experiments validate its feasibility for practical applications.