This paper dedicates to exploring the requirement of a representative volume element (REV) on sample size in the distinct element method (DEM) simulations. Two groups of differently sized dense samples are prepared, respectively, in rigid wall and periodic cells for drained plane strain compression simulations. The results show that, under given boundary conditions, the qualitative consistency of the confinement-dependent shear dilation with experiments, i.e., a higher confining pressure leading to less shear dilation, is observed from adequately large samples, but the qualitative deviation from overly small samples. In the samples of adequate size, the dilation mostly localizes within a shear band, and decrease in the shear band thickness with increasing confining pressure largely results in the confinement-dependent decrease in dilation. In contrast, the overly small samples incapable of completely accommodating a shear band dilate uniformly, and consequently, the uncertainty in the shear band probably brings about the contradictory results. From the standpoint of REV, the criterion that a DEM sample needs to satisfy is that its dimensions need to be "much larger" than the thicknesses of shear bands. Additionally, the formation of shear bands in periodically bounded samples provides strong evidence against the argument of "uniform" deformation patterns. [ABSTRACT FROM AUTHOR]