In this research funded by NSFC (50408038), an agent-based simulation model AMUSE is being developed for the evacuation behaviour of humans induced by the architectural characteristics of the environment. From previous research the architectural characteristics are interpreted as a list of so-called architectural clue types, which are related to three groups of evacuation strategies with a different priority. The Doorway clue is taken as an example of all the clue types to be investigated. With it a basic research method is explained. With an initial six-variable decision making model, a set of virtual scenes were constructed and implemented in a Head-CAVE system, in which 102 subjects were tested as in an evacuation game. With the Binary Logistic Regression the utility function of the model is estimated indicating how these variables affect human choice on any pair of doorways in a scene. Finally, as a sub-model of AMUSE, the decision making model of the Doorway clues is setup, from which evidence was found that the distance from the decision point to the doorway is not always the most important factor as it is assumed in the other evacuation models.