This paper is concerned with the problems and solutions to reliable analysis of arbitrary datasets. Our approach is to describe components of a system called the GARP Modelling System (GMS) which we have developed for automating predictive spatial modelling of the distribution of species of plants and animals. The essence of the system is an underlying generic spatial modelling method which filters out potential sources of errors. The approach is generally applicable however, as the statistical problems arising in arbitrary spatial data analysis potentially apply to any domain. For ease of development, GMS is integrated with the facilities of existing database and visualization tools, and Internet browsers. The GMS is an example of a class of application which has been very successful for providing spatial data analysis in a simple to use way via the Internet.