In order to investigate the impacts of three factors, namely distance, exit density, and passenger flow direction and volume, on passengers’ decision-making during the emergency evacuation of subway stations, data was collected based on a virtual reality experimental platform for subway stations. The conditional Logit model was employed to calibrate the utility coefficients of these three influencing factors. The results indicate that distance, exit density, and passenger flow direction and volume exhibit negative utility values, specifically−0.025, −0.015 and −0.143. Moreover, the negative influence of exit density on the utility of passengers’ exit—selection is less significant compared to that of distance and passenger flow direction and volume. From a quantitative standpoint, this study elucidates the perceptual biases of passengers with respect to influencing factors during the evacuation phase. This offers fresh perspectives for enhancing the accuracy of passenger evacuation prediction models.