The description of sandstone samples, defined by three components, is introduced through non-parametric sample density description. The composition of a particular sample defines a direction from the origin of the Cartesian reference frame. Radial difference between individual samples then provides the measure of their similarity. A sample set can be viewed as a bundle of compositional axes with a corresponding density distribution on both the ternary diagram (or unit constant sum plane) and the positive octant of the unit sphere. Because radial differences are disproportionately displayed on the ternary diagram, radial density calculations on the sphere are to be preferred. The mode of the sample set, the implied centre of greatest data concentration, is considered to be the most meaningful representative value, as the component joint mean can lie outside the data field with certain data configurations. In addition, the component joint mean cannot provide a useful representative composition for samples drawn from a bimodal distribution. Sample density distribution not only provides a data defined approach to describe sample dispersion, but also furnishes an indication of sample sufficiency in displaying the degree to which modes are supported by surrounding observations.