This paper shows that for DEM simulations of triaxial tests using samples with a grading that is representative of a real soil,the sample size significantly influences the observed material response.Four DEM samples with identical initial states were produced:three cylindrical samples bounded by rigid walls and one bounded by a cubical periodic cell.When subjected to triaxial loading,the samples with rigid boundaries were more dilative,stiffer and reached a higher peak stress ratio than the sample enclosed by periodic boundaries.For the rigid-wall samples,dilatancy increased and stiffness decreased with increasing sample size.The periodic sample was effectively homogeneous.The void ratio increased and the contact density decreased close to the rigid walls.This heterogeneity reduced with increasing sample size.The positions of the critical state lines(CSLs) of the overall response in e-log p’ space were sensitive to the sample size,although no difference was observed between their slopes.The critical states of the interior regions of the rigid-wall-bounded samples approached that of the homogeneous periodic sample with increasing sample size.The ultimate strength of the material at the critical state is independent of sample size. This paper shows that for DEM simulations of triaxial tests using samples with a grading that is representative of a real soil, the sample size significantly contributing influences the observed material response. Four samples were identical initially states were produced: three cylindrical samples bounded by rigid walls and one bounded by a cubical periodic cell .When subjected to triaxial loading, the samples with rigid boundaries were more dilative, stiffer and reached a higher peak stress ratio than the sample enclosed by periodic boundaries. For the rigid-wall samples, dilatancy increased and stiffness decreased with increasing sample size.The periodic sample was employed homogeneous ratio. The hetero ratio was reduced compared with increasing sample size. response in e-log p ’space were sensitive to the sample size, although no difference was observed between their slope the critical states of the interior regions of the rigid-wall-bounded samples approached that of the homogeneous periodic sample with increasing sample size. ultimate strength of the material at the critical state is independent of sample size.