This paper presents an efficient time-frequency hybrid method for predicting the tunnel and ground vibrations caused by subway trains. This method contains two steps. Firstly，a 2D multibody system/finite element model of subway train-track-tunnelsoil is established to obtain the track-tunnel interaction forces in the time domain. Secondly，the transformed forces are applied to the 2.5D finite element method-perfectly matched layers model to solve the dynamic response of the tunnel-soil system in the frequency-wavenumber domain. Moreover，to improve the efficiency of frequency-wavenumber domain computation，an efficient wavenumber sampling scheme is proposed. In numerical examples，the accuracy and efficiency of the proposed method are verified by comparing with the time-domain 3D finite element method and the traditional fixed wavenumber domain sampling scheme，respectively. The results show that the time histories of vertical velocity of the observation points calculated by the time-frequency hybrid method are in good agreement with the 3D finite element method. The vertical velocity levels of the observation points calculated by the time-frequency hybrid method are slightly smaller than those by the time-domain 3D finite element method，but the relative error of the two methods is less than 3%. Furthermore，the proposed wavenumber sampling scheme can well predict the effective wavenumber ranges of different frequency loads. In addition to the low-frequency vibration below 2 Hz，the 1/3 octave curves of the vertical velocity calculated by the proposed wavenumber sampling scheme is almost consistent with those by the traditional sampling scheme. In terms of computational efficiency，the CPU time of the proposed wavenumber sampling scheme is 3/5 of that of the fixed wavenumber domain sampling scheme.