Prior to resource modelling, sample scale data is normally grouped into spatial domains that reflect zones of homogenous properties. These domains are usually based on a combination of grade and other variables that reflect, for example, geology and alteration, or variables known to correlate with properties of the samples that affect downstream processing. Domaining and wire-framing sample scale data is a time intensive task and it is difficult to track more than a few variables simultaneously while establishing domain boundaries. This is especially the case in operations with complex data, for example, a combination of multi-element assaying of total element concentrations, and/or phase specific assaying such as sulfide sulfur, carbonate carbon, cyanide available gold, and/or downhole geophysical measurements. To minimise unintentionally grouping samples of contrasting properties within the same domain, unsupervised statistical methods are available that can automatically identify samples atypical of the domain, both as univariate tests, and as tests that take correlation within the domain into account. Unsupervised methods in this context means methods of identifying atypical samples that require just the selection of the domain and input variables. Thresholds are derived robustly from the data itself, and the results are therefore repeatable and operator independent. Atypical samples may be thought of as outliers, ie, samples that are separated from the majority' of the data. If these atypical samples are themselves spatially associated, they may be allocated to an alternative existing domain or allocated to a new domain. Care should always be taken to ensure that identified outliers are due to meaningful causes and not due to data coding and storage errors, although the identification of the latter is still valuable._x000D_
*This is an abstract only. No paper was prepared for this abstract.*