The majority of SAG and ball mill design methodologies are applied using a single design value for mill selection, such as the 80th percentile ball mill work index or SMC value. Inherent in the selection of this design value is a desire for the mill to accommodate unfavourable, but not absolute worst case, feed conditions. It has been observed in numerous post-mortems of unsuccessful designs and also in many optimisation studies, that the selection of the design value has been poor. Statistical analysis of testwork data sets shows that inherent variability in the data can, by itself, lead to design errors. Metallurgical testing is an attempt to measure the characteristics of millions of tonnes of ore using only a few kilograms of sample. In this paper some realistic but synthetic test work data sets are used to explore old and new methods of defining the design envelope and subsequently selecting a reliable point. The aim of the paper is to provide practitioners with tools that reveal the adequacy (or otherwise) of metallurgical test programs intended to support various levels of project design. It also provide guidance as to how it is possible to statistically justify a comparatively high design point in the absence of adequate test data.CITATION:David, D, 2015. Measuring and taking notice of orebody variability - an essential ingredient for reliable plant design, in Proceedings MetPlant 2015, pp 10-26 (The Australasian Institute of Mining and Metallurgy: Melbourne).