Traditional　ground　motion　theoretical　models　in　time　domain　or　frequency　domain　can＇t　fully　represent　the　nonstationarity.　The　time-varying　power　spectrum　or　evolutionary　spectrum　for　ground　motion　can　be　calculated　by　the　modern　signal　analysis　methods.　However,　the　corresponding　accuracy　evaluation　criterion　is　inadequate　at　present.　The　wavelet　packet　decomposition　method　is　verified　as　an　efficient　tool　in　establishing　time-varying　power　spectrum　on　the　basis　of　theoretical　and　computational　analysis.　The　accuracy　evaluation　criterion　is　presented　based　on　the　marginal　condition　in　time-frequency　domain.　The　time-frequency　characteristics　in　ground　motions　are　studied　on　the　basis　of　signal　process　and　statistical　analysis,　and　the　time-frequency　spectrum　are　resolved　into　the　multi-peak　smooth　spectrum　and　the　non-stationary　noise　spectrum.　The　frequency　domain　spectrum　model　which　includes　multi-peak　smooth　power　spectrum　and　non-Gaussian　noise　spectrum　in　frequency　domain　is　established.　The　time　domain　spectrum　model　which　includes　multi-peak　smooth　energy　spectrum　and　non-Gaussian　noise　spectrum　in　time　domain　is　established.　The　time-frequency　spectrum　embodies　the　unity　in　the　aspects　such　as　characteristics　both　in　time　domain　and　frequency　domain,　functions　expression,　accuracy　and　multidimensional　analysis　of　ground　motion.　The　good　precision　and　strong　nonstationarity　can　be　obtained　if　the　natural　earthquake　wave　is　adjusted　or　the　artificial　ground　motion　is　generated　according　to　the　unified　time-frequency　spectrum.　©　2016,　Nanjing　University　of　Aeronautics　an　Astronautics.　All　right　reserved.