The purpose of this paper is to argue for a novel use of geographical information system (GIS) as an exploratory device for understanding complex space-time processes. Conventionally, a GIS has been configured as a spatial database management system. Therefore, the capabilities of data input, storage, retrieval, manipulation and display have been well developed, whereas the analytical and dynamic processing and modelling capabilities are under-developed. The importance of integrating GIS with dynamic and analytical models has been widely recognised. Extending the use of GIS into spatial simulation represents a plausible research direction. On the other hand, the usefulness of conventional deductive model becomes limited when dealing with complex space-time processes. In essence, the inherent complexity calls for a simulation approach. The new nonlinear modelling paradigm, for example, cellular automata (CA), opened a way in which behaviourally-richer dynamics and more micro-spatial data could be incorporated into understanding space-time processes. It is argued therefore that GIS can be used as a platform - a controlled environment or laboratory - for exploring complex space-time processes. The discussion of GIS-based simulation in this paper is illustrated with a primitive simulation of the evolution of urban spatial structure.