The primary purpose of road lighting is to satisfy the visual requirements of road users, and then to reduce the energy consumption as possible. Moreover, these two factors are closely related to the light intensity spatial distribution of road luminaires. Only one type of light distribution is not applicable to different roads and installation conditions of road luminaires. In view of three typical roads as major road, collector road and local road, a reverse method is proposed to design the most energy-saving light distribution of LED road luminaires based on the specific road conditions, installation parameters of road luminaires and practical visual requirements for drivers. A nonlinear optimization model is established, which takes the visual requirements of drivers as the constraint conditions and the minimum total luminous flux of LED luminaires as the goal. Then a polynomial of cosine function for the illumination distribution on road is used, and a two-level optimization strategy is proposed for the nonlinear optimization. Finally, the most energy-saving light distributions are obtained at three different road conditions. Compared with the light distribution of LED road luminaires designed by equal illuminance distribution method, the LED road luminaires with the proposed nonlinear method can significantly improve the quality of visual lighting environment on the road and save energy by about 30%.