Statistical frequency distributions may be used to characterize the overall variation of grade in orebodies. Special types of distribution (e.g. lognormal, log-type A) are necessary to describe skewness inthe case of rare metals. Also needed (but for the most part hitherto neglected) is the serial correlation function which describes the tendency for grades near together in an orebody to be closer together than grades at points relatively far apart.The sampling problem for orebodies is interpreted in statistical terms. Different methods are reviewed, and the effect of serial correlation discussed.A discussion is given, based on experiments and observations made by the author, of biases arising in the minei:l material by such methods as chip sampling and grab sampling due to such causes as nonrepresentative selection by the sampler.INTRODUCTIONMiners have always recognized the large element of chance in their operations. Chance is the raw material of statistical theory and, provided there is some recognizable pattern or underlying regularity in a set of data arising from the operation of chance events, statisticians are able to formulate mathematical models to study them. Mining engineers, geologists, and statisticians have gone a long way between them towards finding general statistical rules which describe the distribution of minerals and metals in orebodies, and even the distribution of orebodies over the earth's surface.These rules do not have the force of physical laws, though they obviously derive from them and have been used to throw light on them. From statistical theory it is possible to find methods of estimating from sampling some of the characteristics of orebodies which are of importance to...