This paper introduces GIS-based statistical methods for the study of settlement patterning in an archaeological context. The main aspect is an improvement over suboptimal stepwise statistical model selection procedures, with the introduction of a Deletion/Substitution/Addition (DSA) algorithm, a systematic model selection algorithm that provides an optimal model choice. The paper also introduces the combined use of Receiving Operator Characteristic (ROC) curves, and variable ranking as key tools aimed at an understanding of settlement patterning causation. An illustrative case study is provided whereby both a purely environmental and a mixed environmental/historical model (based on settlement hierarchy considerations) are investigated. © 2011 Elsevier Ltd.