An overview over the preparation of representative samples from bulk materials is given. Special attention is being paid to the statistics of subsampling with use of spinning rifflers. A model is provided for the description of a riffling process. The model takes equipment-specific factors (velocity of the riffler, accuracy of the feeder/riffler) as well as material specific factors into account. It is shown that the relative standard deviation of trace species in coal is far higher than those for ＇＇bulk＇＇ properties, such as ash content. The effect of clustering of particles is an increase in over-all standard deviation. The model is also applied to cross-riffling. These calculations showed the empirically observed fact, that a subsequent cross-riffling does not decrease the standard deviation in the samples. After one cross-riffling step, the standard deviation of the subsamples is at a specific value, which cannot be improved by repeating the process.