A reasonable seismic capacity model is crucial for establishing the seismic performance level system and evaluating the seismic reliability of subway station structures. However, the deterministic structural and geotechnical mechanical parameters are usually applied to calibrate the seismic performance levels of subway station structures in the traditional seismic capacity analysis, which ignores the stochasticity of the soil-subway station interaction system. To overcome the challenge caused by the stochastic interaction system, the probability space partition method and stochastic pushover analysis method are combined to develop a calibration strategy of seismic performance levels considering the complete probabilistic information of the stochastic interaction system, and the non-parametric probabilistic seismic capacity models of the subway station structure are further established based on the principle of probability conservation in this paper. A subway station is also taken as the prototype to investigate the applicability of the proposed strategy and the influence of system randomness on the seismic capacity of the subway station structure. The results demonstrate that the seismic performance levels calibrated according to the proposed strategy can effectively consider the complete probabilistic information of the interaction system, which are more rigorous than the existing performance levels. Meanwhile, the probability density evolution of the bearing capacity of the subway station structure is essentially a non-stationary stochastic process, and the non-parametric probability density curves of seismic capacity display noticeable multi-peak characteristic. Moreover, the seismic capacity for LP1 and LP2 levels is more sensitive to the variability of geotechnical parameters above and below the structure, while the former for LP3 and LP4 levels is more sensitive to that on both sides of the structure. The relevant conclusions can provide some guidance for seismic design and improvement of the performance limits of underground structures in the related codes.