Remediation of sites contaminated by unexploded ordnance is complicated by the problem of discriminating between buried conductors that are intact munitions and those that are harmless scrap metal. Here we present two distinct approaches of object discrimination, both of which rely on training data in the form of polarizability curves. These curves show remarkable similarity for same-type objects over a broad range of depths and attitudes, but marked differences when comparing curves from different object types. The first method, called the “voting scheme,” compares field data polarizabilities against templates in a series of cross validations. The second method applies Bayesian statistics on features extracted from the polarizabilities. Here the methods are applied to a 346 element dataset. The voting scheme misclassified fewer UXO objects, but at a cost of more false digs. A hybrid technique combining both methods generates an ordered dig list that ensures efficient use of cleanup resources. For the dataset considered here, the hybrid method identifies over 80% of UXO before the first hole containing scrap metal is dug, with only 7 false digs before all 219 UXO are excavated.