Most existing stress analyses of the skeleton which consider cancellous bone assume that it can be modelled as a continuum. In this paper we develop a criterion for the validity of this assumption. The limitations of the continuum assumption appear in two areas: near biologic interfaces, and in areas of large stress gradients. These limitations are explored using a probabilistic line scanning model for density measurement, resulting in an estimate of density accuracy as a function of line length which is experimentally verified. Within three to five trabeculae of an interface, a continuum model is suspect. When results as predicted using continuum analyses vary by more than 20-30% over a distance spanning three to five trabeculae, the results are suspect.