This article proposes a time-domain procedure for a non-Gaussian stationary random vibration test with prescribed power spectral densities. Previous procedures for generating non-Gaussianity suffered from certain defects. For example, zero-memory nonlinear transformation, an algorithm frequently applied to transform Gaussian signals into non-Gaussian signals, often produces changes in both auto-power spectral densities and cross-power spectral densities, which might result in control instability under certain circumstances. In this article, the authors propose a different approach for the zero-memory nonlinear function. First, a time-domain procedure for a non-Gaussian random test is introduced. Second, a rescaling method is applied to correct the magnitude amplification on the auto-power spectral density because of zero-memory nonlinear transformation. We offer experience formulas in this method to adjust the auto-power spectral density of both super-Gaussian and sub-Gaussian responses. Third, a control strategy using a finite impulse response filter is proposed to further improve the auto-power spectral density if the shape of the auto-power spectral density is distorted. The kurtosis loss induced by the filtering process is also analysed and a corresponding solution is put forward to ease the reduction. Numerical test and a biaxial shaker table test are conducted to validate the availability and superiority of the proposed procedure.