This paper reports on a molecular simulation study of intrusion and retraction of a non wetting fluid in silica nanopores without or with morphological defects (constrictions or undulations) to mimic mercury porosimetry in porous materials. All the pores considered in this work are of a finite length and are connected to bulk reservoirs so that they mimic real materials for which the confined fluid is always in contact with the external phase. Our results for pores with constant or variable cross-section are in qualitative agreement with experimental data: (1) the intrusion and retraction pressures are larger than the bulk saturating vapor pressures, (2) the intrusion and retraction pressures are decreasing functions of the pore size, and (3) entrapment of the confined fluid leads to irreversible intrusion/retraction isotherms (except for the smallest pore for which intrusion is reversible). Our data show that the intrusion pressure varies linearly with the pore diameter when both quantities are plotted in a logarithmic scale; the slope, -1.1 +/- 0.1, is close to the theoretical value -1 given by the Washburn – Laplace equation (the contact angle Θa is found about 102° +/- 15°). In contrast, the retraction pressure varies with a larger slope -1.6+/- 0.1, due to the fact that retraction consists of the nucleation of a gas bubble within the pore. We find that the intrusion chemical potential for the large cavities of the constricted pores corresponds to that for a regular cylindrical nanopore having the same diameter than the constriction, i.e. intrusion in the constricted pores occurs when the non wetting fluid invades the constrictions that isolate the large cavity from the bulk external phase. On the other hand, the retraction chemical potentials for pores with uneven diameter underestimates the value found for a regular cylindrical pore having the pore diameter of the wider parts of the variable-diameter pore. These results suggest that intrusion and retraction experiments (porosimetry) can be used to assess and characterize morphological defects in nanopores, provided that the pore size is known from other independent measurements. The present work also provides a theoretical frame to explain some discrepancies observed between properties assessed by mercury porosimetry and nitrogen adsorption.