It is known that there is stress concentration at specimen ends during experimental rock tests. This causes a deviation of the measured nominal (average) stresses and strains from the actual ones, but it is not completely clear how strong it could be We investigate this issue by numerical modeling of the hydrostatic tests using a reasonably simple constitutive model that reproduces the principal features of the behavior of porous rocks at high confining pressure, P-c. The model setup includes the stiff (steel) platens and the cylindrical model rock specimen separated from the platens by the frictional interfaces with friction angle phi(int). The whole model is subjected to the quasi-statically increasing normal stress P-c. During this process, the hydrostats P-c(epsilon) are computed in the same way as in the real tests (epsilon is the average volume strain). The numerical hydrostats are very similar to the real ones and are practically insensitive to phi(int). On the contrary the stresses and strains within the specimen, are extremely sensitive to phi(int). They are very heterogeneous and are characterized by a strong (proportional to phi(int)) along-axis gradient, which evolves with deformation. A strong deviation of the stress state at the specimen ends from the isotropic state results in inelastic deformation there at early loading stages. It follows that the nominal stresses and strains measured in the experimental tests can be very different from the actual ones, but they can be used to calibrate constitutive models via numerical simulations. (C) 2015 Elsevier Ltd. All rights reserved