Abstract: To avoid the epistemic uncertainty of seismic fragility of subway station structures caused by the incomplete seismic demand samples, a novel procedure of seismic fragility analysis of subway station structures considering the statistical uncertainty of seismic demands is proposed. The limited set of seismic demand samples is firstly converted into a large sample problem based on the Bootstrap method. Then, the joint probability distribution model of statistical uncertainty variables is further established by combining the maximum entropy principle and copula theory. On this basis, the variability of seismic fragility of subway station structures caused by the statistical uncertainty of seismic demands is further quantified, and the mean fragility curves and envelope fragility curves with certain confidence level are obtained. Finally, the Daikai subway station is taken as the prototype to investigate the influences of the statistical uncertainty. The results show that the seismic fragility of subway station structure derived from the limited seismic demand samples has significant variability which shows a tendency of increase and then decrease with the increase of ground motion intensity. Moreover, the mean fragility curves and envelope fragility curves can effectively reflect the uncertainty of seismic fragility derived from the limited seismic demand samples, and have higher reliability. It may provide reference and guidance for the seismic performance and seismic risk assessment of subway station structures.