To address the problems of complex and rich details of face features and heavy work and high cost of reconstructing high-precision three-dimensional (3D) face models, a high-precision 3D face reconstruction algorithm based on the gradient light image is proposed. Six different illumination modes are established using a spherical light device, and specular and diffuse reflection lights are separated using the polarization characteristic. Using a unidirectional gradient light image, specular and diffuse normal maps are obtained by calculation based on relationship between the reflection and normal directions by employing the bidirectional reflectance distribution function (BRDF). The obtained maps are merged and mapped to the low-precision mode to reconstruct a high-precision 3D face model. The experimental results show that the proposed method has low computational complexity and high reconstruction accuracy. Based on the obtained results, it is best to use the high-pass filtering with a window size of 7×7 enhanced specular and diffuse reflection normal maps to achieve the best reconstruction at 2∶1 map ratio fusion. The proposed model can be refined to microscale.