Agent-based modeling (ABM) is currently a new active research area in Geocomputation. The methodology of ABM is integration of the theories and technologies of complex adaptive system, artificial life and distributed artificial intelligence. The complex system such as geographical system is conceived as societies of autonomous agents that are able to act both on themselves and on their environments. The agents can communicate and interact with other agents. The determinants of an agent's behavior have a local character and there is no global constraint on the system's evolution. Therefore, it is a good alternative way of simulating the evolutional process of the spatial structure by modeling behaviors of these local active agents and their interactions. Such a multi-agent model allows a greater variety of spatial interaction, including variable extension of the spatial range of interactions, which can be defined by the connectivity of a network according to the characteristics of each agent. Moreover, instead of allowing only a few quantitative variables in non-linear equation and possible states for each cells in cellular automata, the ABM is able to integrate any qualitative or quantitative description of an agent, whose behavior may be very complicated. The flexible modeling method allows for a much more detailed representation of spatial interactions and of some local properties and also makes it possible to introduce new agents or new rules in the model without changing the other parts. This paper gives an overview on the theory background and technology advantages of ABM compared with Equation-based Modeling(EBM) and Cellular Automata(CA). Since interaction and adaptation between agents is the central task of ABM, this article gives a detail discussion on the structure, characteristics and internal mechanisms of an agent. Its related problems such as research advance and the platform of ABM are also surveyed.