Cultural resource databases represent the single largest compilations of archaeological site data, but these databases are seldom used in research because they were designed for management purposes, evolved from paper-based inventories, contain significant interobserver variation, and record information inconsistently. In this paper we present methods designed to alleviate these problems in an analysis of more than 3,000 ancestral Pueblo habitation sites from southwestern Colorado. Our methods draw heavily upon Bayesian statistical concepts and utilize the rich excavation records of our study area to quantify the relationship between surface evidence and excavation results using probabilities. This approach offers a number of advantages over ad hoc, judgmental approaches, and produces a more empirically justified history of ancestral Pueblo settlement in our study area. We believe methods like these have great potential for reconstructing settlement patterns from survey data.