Rapid advances in computing technology have facilitated the storage and manipulation of large sets of data in Geographic Information Systems. These data present new opportunities for modelling spatial distribution and change to distribution at a regional scale. The models used should take account of the special nature of GIS data, particularly the differing resolution associated with different variables. Likewise, inferences, either concerning model selection or model predictions, should be made with reference to the sampling units on which the estimates are based. Computer-intensive statistical methods, especially the bootstrap, are suitable for studying the variability of predictions. Presentation of results in map form should be accompanied by estimates of error.