During a fire evacuation, long lateral evacuation distances, large crowds waiting for evacuation at the same level, and easily panicked populations are common. This research aimed to look into the large-scale evacuation behavior of urban underground complexes with limited evacuation and egress during a fire. A simplified model for large-scale group evacuation of urban subsurface complexes was constructed using system dynamics theory. The Vensim software was used for quantitative simulation. The model could represent the typical phenomenon of group evacuation behaviors, such as quick or slow, under seven operating situations with total initial numbers of 350, 400, 450, 500, 1000, 2000, and 4000. The results of an analysis of critical affecting factors show “total initial number” and “panic state” during a large-scale group evacuation: a large beginning population will result in a rapid reduction in system evacuation capability, delaying the completion of the evacuation process significantly; meanwhile, if the level of panic is deficient, the system’s evacuation efficiency will remain low for an extended period, making it difficult to evacuate trapped persons promptly. According to the findings, the developed system dynamics model, which combines the advantages of a continuous model with the advantages of a discrete model, is very accurate. At the same time, we should emphasize the importance of the evacuation guide and reinforce the fire education and behavior drills for the building’s workers. This research presents a simplified model for the evacuation of large groups of people from metropolitan underground complexes. Furthermore, the findings may give theoretical support for the development of rules and safety management practices.