Abstract: In performance-based structural seismic design, the probabilistic seismic demand analysis model assumes a regression analysis model to correlate seismic intensity measure with engineering demand parameters. An appropriate regression analysis model can more accurately reflect the relationship between seismic intensity measure and engineering demand parameters. To study the regression analysis model of seismic intensity measure applicable to subway station structure, a two-dimensional finite element analysis model of soil-subway station interaction was developed based on the integral nonlinear dynamic time history analysis method, and ground motion with velocity pulse was selected as the excitation input. This study analyzed the influence of unitary linear regression, log-domain unitary linear regression and log-domain binary linear regression analysis models on the optimal seismic intensity measure selection of subway station structure. The results show that the unitary linear regression analysis model is not suitable for analyzing subway station structure under lateral seismic excitation. Instead, the log-domain unitary linear regression analysis model is more appropriate. While the log-domain binary linear regression analysis model improves the goodness-of-fit, the increase in the coefficient of determination is not substantial. Therefore, it is suggested to adopt the log-domain unitary linear regression analysis model in the probabilistic seismic demand analysis of subway stations structure.