AbstractAn important property of loess is a tendency to collapse on loading and wetting (hydroconsolidation) which can have serious consequences worldwide for civil engineering projects. This paper describes the use of Monte Carlo and other analytical techniques to predict the shape of naturally occurring loess particles. Randomly generated particles are classified according to Zingg shape categories: disc, sphere, blade and rod. By assuming a uniform distribution for the basic particle, average relative dimensions are calculated for the blade category, into which most loess particles have been shown to fall.