Recent research on the size distribution of diamonds from micro-size (0.1 to 0.5 mm) to macro-size (larger than 0.5 mm) in kimberlites and lamproites has shown that the size distribution is continuous. As a consequence, provided the right statistical distribution model is used, the grade of macro-diamonds can be estimated by extrapolating the size distribution of the micro-diamonds. The lognormal size distribution of the micro- and macro-diamonds is an artefact of the sampling recovery process. The underlying complete size distribution is supposed to be a loghyperbolic distribution, obtained by mixing lognormal size distributions, corresponding to different diamond growth processes. The extrapolated grade estimates based on the lognormal model are only approximate, but are of much practical use in ranking targets for bulk sampling or in verifying grade continuity at depth. The value distribution of diamond in kimberlite is best approximated with the lognormal model, In such case, the t-estimator can be used instead of arithmetic mean to determine the average carat price, by dividing the t-estimator of the stone values by the t-estimator of the stone sizes. However, the heavy tail of fine and poor quality diamonds at the lower end can create deviations from lognormality and a bias in the t-estimator. The deviations can be corrected by applying three-parameter lognormal distributions or by calculating the parameters from the weight frequency distribution, which often gives a better fit to the lognormal model. than the number frequency distribution. Extreme value graphs of the value distributions show that some diamond populations in kimberlite have no steady mean: the larger the sample, the more likely the average value will increase.
