Abstract: The selection of input motion is of most importance in seismic design and analysis of underground structures. In recent years, the concept of “the most unfavorable ground motion” has been widely studied and applied in the field of seismic design for surface structures. However, there is no research on ranking the seismic input motion based on the seismic response of underground structures. To explore the applicability of the concept of “the most unfavorable ground motion” for the seismic design of underground structures, a circular tunnel in soft soil is taken as the research object, and the corresponding soil-structure dynamic analysis modelis established. Taking the ovaling deformation rate of the circular liner as the critical index of tunnel responses, the distribution as well as the difference of dynamic responses of the tunnel structure under excitations of 18 different input motions are studied with the dynamic time-history method, and thus the ranking of seismic input motions based on the tunnel response is obtained. By means of amplitude modulation, the effects of peak ground acceleration （PGA）and tunnel depth on the input motion ranking are further investigated. Finally , the correlation between different input motion parameters, including peak acceleration, peak velocity, peak displacement, absolute cumulative velocity （CAV） and Arias intensity （I_A）, and the tunnel seismic response is evaluated. Results show that, ① When PGA increases from 0.5 m/s² to 2 m/s², the ranking of ground motion changes apparently, and the diversity of tunnel responses caused by different input motions is significantly amplified, such that the ratio of the maximum diametrical deformation rate caused by the most unfavorable input motion to the mean value is increasing from 1.1 to 1.9; ② The input motion ranking results remain unchanged for different depths of the tunnel; ③ For the PGA of 2 m/s², the correlation between the tunnel response and the peak bedrock velocity （PBV） matches well, i.e. the correlation coefficient is 0.94, then followed by the peak bedrock displacement （PBD） and the I_A, i.e. the correlation coefficient is 0.62 and 0.48, respectively, and the worse is, the peak bedrock acceleration （PBA） and the CAV have a correlation coefficient of only 0.37 and 0.22, respectively. The conclusions provide a scientific basis for the selection of seismic input motions of soft soil tunnels.