Pedestrian exit selection behavior during evacuation process in a rail transit station is restricted by pedestrians’ perceptual as well as cognitive abilities. Moreover, pedestrian may reselect the exit according to his or her current position and the realtime exit conditions. To simulate these behaviors, a dynamic exit selection model is presented. A perceptual parameter and a series of cognitive coefficients are introduced in the model to reflect pedestrians’ perceptual and cognitive level. Besides, two subjective evacuation time computing approaches are proposed to simulate pedestrians’ exit selection and reselection behavior. A scenario of a station platform is built. A number of simulation experiments are run to study the effect of pedestrians’ perceptual parameter and cognitive coefficients on total evacuation time. The results show that the total evacuation time is sensitive to model parameters, furthermore, exit selection and reselection behaviors can be modeled. It is found that when pedestrians can perceive only a few exits or pedestrians prefer waiting, the station staff should guide pedestrians who evacuate from the crowded exit to use the unobstructed exit; whereas when pedestrians can perceive enough exits and they prefer walking, measures should be taken to help them to choose a rational exit and let them wait a bit longer in front of the exit.