Hypergeometric statistics have recently been used to establish a quantitative measure of performance for geochemical exploration techniques over known mineral occurrences. Using this method, the effectiveness of new exploration techniques can be compared objectively with conventional approaches. Sample site classification is the basis on which survey results are compared. This performance measure requires prior knowledge of the location of mineralization and a model for element dispersion from the primary into the secondary environment. This information allows assignment of sample sites that should give ‘anomalous’ (e.g. those overlying mineralized zones) or ‘background’ results to be identified prior to the orientation survey. Previous application of hypergeometric statistics requires that a high contrast exists between ‘anomalous’ and ‘background’ subpopulations in the geochemical orientation data so that there is no uncertainty in the classification of samples. In this paper, a refinement is developed that allows consideration of geochemical variables that do not exhibit this required high level of geochemical contrast, that is, where ‘anomalous’ and ‘background’ subpopulations exhibit significant overlap. This refinement involves determining the hypergeometric probabilities of obtaining the same result at random (P(x)) for a range of geochemical thresholds, instead of only one threshold (i.e. the one used to classify anomalous and background samples in cases with high geochemical contrast). At each threshold, different numbers of samples will be classified as ‘anomalous’ and different numbers of ‘anomalous’ samples will occur at ‘anomalous’ sites. As a consequence, the resulting random hypergeometric probabilities will change with threshold level. Using a range of thresholds to classify the geochemical orientation survey results allows identification of the minimum hypergeometric probability (MHP) for the dataset. Using this threshold, the classification of anomalies will bear the least resemblance to what would be expected if the survey results were generated at random. Employing this refined MHP approach, one can simultaneously evaluate the effectiveness of an exploration method, and select the threshold that optimally classifies anomalous and background samples.