The three typical applications of G I S i n archaeology have been analysis or, more rightly, visualization; management; and the development of "predictive" models . A predictive model is defined as "hypotheses or sets of hypotheses which simplify complex observations whilst offering a l argely accurate predictive framework struc­ turing these observations" (Clarke 1 968 : 32) . Correlative predictive models are those that "identify and quantify relationships between archaeological site locations and environmenta l variables" ( Sebastian and Judge 1 98 8 : 4) . Explanatory predictive models are "models that are deductively derived and attempt to predict hmv particular patterns of human land usc will be reflected in the archaeological record" (Sebastian and Judge 1 98 8 : 4) . We would argue that except for management appli cations, the vast majority of GIS studies have confused either "pretty pictures" with innovation, as in the case of visualization, or the statement of simple cOlTelations with theory, as in the case of predictive models . This is not to say that previous predictive modeling efforts were wasted; indeed, without these early studies many of the tools discussed

here would have remained undeveloped. However, predictive modeling in archae­ ology has entered a period of doldrums that will be overcome only when we shift from a methods orientation to a theoretical one . GI S grew out of efforts by land managers to construct a visually linked database that would allow them to track growth, would provide a basis for planning, and would allow them to isolate areas suitable for specific activities or facilities . Geographers and biologists then began to use GIS to conduct spatial analysis of various activities or populations . With the growth of cultural resource regulations in the 1 970s coupled with large-scale impact projects (e .g . , strip mines), land managers were immediately interested in the promise of predictive modeling. Predictive modeling efforts developed out of a desire on the partoflandmanagerstocategorizetheirlandsastothelikelihoodofsitepresence. Therewasaninitialhopethatlandsplacedinalow-likelihoodcategorycouldbe essentiallyexemptedfromfurtherinvestigation.Thisattitudewasquicklytempered intothepresentgoalofflaggingareaswithahighlikelihoodforsitepresenceasan aid in management decisions rather than as a way to exempt lands .