In the last decade, a lot of research activity took place to unveil the properties of granular materials 1,2 , primarily because of their industrial importance, but also due to their fascinating properties. This has unraveled many interesting and so far unresolved phenomena (for example, clustering, size-segregation, avalanches, the coexistence of gas, liquid and solid, etc.). Under highly excited conditions, granular materials behave as a fluid, with prominent non-Newtonian properties, like the normal stress differences 3 . While the normal stress differences are of infinitesimal magnitudes in a simple fluid (e.g. air and water), they can be of the order of its isotropic pressure in a dilute granular gas 4 . From the modelling viewpoint, the presence of such large normal-stress differences readily calls for higher-order constitutive models 5,6 even at the minimal level.Studying the non-Newtonian behaviour is itself an important issue, since the normal stresses are known to be the progenitors of many interesting and unique flow-features (e.g. rod-climbing or Weissenberg-effect, die-swelling, secondary flows, etc. 7 ) in non-Newtonian fluids. Also, normal stresses can support additional instability modes (for example, in polymeric fluids and suspensions 7−10 , which might, in turn, explain some flow-features of granular fluids. For example, particle-clustering 11−13 has recently been explained from the instability-viewpoint using the standard Newtonian model for the stress tensor 12,14,15 The kinetic theory of Jenkins & Richman 16 first showed that the anisotropy in the second moment of the fluctuation velocities, due to the inelasticity of particle collisions, is responsible for such normal stress behaviour. They predicted that the first normal stress difference (defined as N 1 = (Π xx − Π yy )/p, where Π xx and Π yy are the streamwise and the transverse components of the stress deviator, respectively, and p is the isotropic pressure, see section IIB) is maximum in the dilute limit, decreases in magnitude with density, and eventually approaches zero in the dense limit. Goldhirsch & Sela 4 later showed that the normal stress differences appear only at the Burnett-order-description of the Chapman-Enskog expansion of the Boltzmann equation. Their work has clearly established that the origin of this effect (in the dilute limit) is universal in both atomic and granular fluids, with inelasticity playing the role of a magnifier and thus making it a sizeable effect in granular fluids. While the source of the normal stress differences in the dilute limit has been elucidated both theoretically and by simulation, its dense counterpart has not received similar attention so far. This is an important limit since the onset of dilatancy (volume expansion due to shear 17,18 ), crystallization, etc. occur in the dense regime, which in turn would influence the normal stress differences.Previous hard-sphere simulations 19,3,11 did look at the normal stress differences, but they did not probe the dense limit in a systematic way. The...