Abstract: In this paper we describe the design and development of an interactive and generic agent based simulator, providing valuable support for organizing the emergency rescue plans of a large-scale accident. Analysis of real rescue activities has been conducted in collaboration with medical experts in order to understand the collaborative process and the involved actors and features. Based on the emergency analysis, an agent-based model and simulator was constructed including (1) the autonomous Agents – representing victims with evolving illness and rescuers (doctors, nurses, fireman) collaborating to rescue the first ones; (2) the Environment -representing the accident site having obstacles and dangerous areas and where the victims are initially spread and the doctors move to explore -perceive – treat and helpers evacuate; (3) the Interactions between rescuers – exploring collectively, evacuating by pairs, communicating directly or via artefacts- (4) the Organization of actors as distributed “independent” sub-teams in various site sub-zones or as a centralized whole team conducted by the rescue chief; and (5) the User interfaces allowing mainly initial configuration of the simulations (e.g. number of victims and states, followed strategies, rescuers behaviours), continuous visual control of the process of rescuing (e.g. site overview with acting-interacting agents, graphics, text descriptions), dynamic changes of parameters of an on-going simulation (e.g. adding new victims, adding new rescuers, or adding dangerous zones or new obstacles on sites) as well as step-by-step simulation. This simulation shows that it is possible to create a virtual environment with cooperating agents interacting in a dynamic environment. On-line and off-line analysis of simulation traces and results enable us first understanding complex situations in rescuing activities in large-scale accidents, and than planning for responding to crisis situation. This simulation approach is useful for identifying the best scenarios and eliminating potential catastrophic combinations of parameters and values, where rescue performance could be significantly impacted. © Proceedings ISCRAM 2004.