Wilson-θmethod is a kind of the numerical method for structure dynamic response, and it can guarantee the unconditional stability under certain conditions. Like other identification method in time domain, Wilson-θinverse analysis method for load identification, also has the problems of accumulated error and accuracy. The paper studies the inverse analysis of Wilson-θ method and some influence factors for load identification precision, including the value of θ, the time interval, damping ratio, the number and position of measurement points and the noise. Numerical examples are implemented to investigate each parameter impact on the load identification precision, in order to provide references on the choice of the parameters for high-precision load identification. To improve the load identification accuracy problem caused by the cumulative error, the golden section is introduced to modify the inverse method of Wilson-θ. Simulation examples show that the modified algorithm is stable and has good noise-resistance capability.