This paper presents a development of the extended Cellular Automata (CA), based on relational databases (RDB), to model dynamic interactions among spatial objects. The integration of Geographical Information System (GIS) and CA has the great advantage of simulating geographical processes. But standard CA has some restrictions in cellular shape and neighbourhood and neighbour rules, which restrict the CA’s ability to simulate complex, real world environments. This paper discusses a cell’s spatial relation based on the spatial object’s geometrical and non-geometrical characteristics, and extends the cell’s neighbour definition, and considers that the cell’s neighbour lies in the forms of not only spatial adjacency but also attribute correlation. This paper then puts forward that spatial relations between two different cells can be divided into three types, including spatial adjacency, neighbourhood and complicated separation. Based on traditional ideas, it is impossible to settle CA’s restrictions completely. RDB-based CA is an academic experiment, in which some fields are designed to describe the essential information needed to define and select a cell’s neighbour. The culture innovation diffusion system has multiple forms of space diffusion and inherited characteristics that the RDB-based CA is capable of simulating more effectively. Finally this paper details a successful case study on the diffusion of fashion wear trends. Compared to the original CA, the RDB-based CA is a more natural and efficient representation of human knowledge over space, and is an effective tool in simulating complex systems that have multiple forms of spatial diffusion.