In order to meet the needs of virtual experiment in engineering, a 3D stochastic road roughness modeling method using power spectral density function is studied. By using the road spectrum fitting formula of power function adopted in the national standard, the analytical expression of two-dimensional power spectrum of road surface is derived, which avoids the parameter estimation problem of rational function method. The process of generating 3D stochastic pavement by two-dimensional inverse Fourier transform is established, and the E-grade pavement in the national standard is simulated by the performance requirements of a certain automobile vibration system. The power spectrum of the simulated pavement is estimated by using periodogram method and AR parameter model method respectively. The estimated results show that the simulated pavement spectrum is highly consistent with the standard pavement spectrum. The method based on two-dimensional inverse discrete Fourier transform is not only accurate and fast, but also has strong practicability.