A method is established to generate pore networks within domains of arbitrary shape, as long as the domain can be mathematically described by a set of inequalities. In this method, a stochastic network algorithm is adopted to construct pore network skeletons, which are then cut into the desired shapes using a new pore network cutting algorithm. The latter can be embedded into other methods to transplant its 'pore network cutting' function. Using this method, pore networks with four archetypical two-dimensional shapes (namely, cross-sections of one-holed rings, trilobes, four-holed rings, and wheels) and four three-dimensional shapes (namely, spheres, cylinders, trilobes, and hollow cylinders) are constructed as examples. Then, some of these pore networks are applied to simulate diffusion and reaction in Pd/c-alumina catalyst particles for hydrogenation of benzene to cyclohexane. It is shown that the randomness of the pore network and the external particle shape significantly affect the performance of catalysts, because of their impact on effective diffusivity and diffusion length, respectively, indicating that this structural information must be accounted for to achieve a model with high accuracy. The versatile method proposed in this article is ideal to study the effect of particle shape and pore network structure on the performance of porous materials for catalysis and other applications. (C) 2017 Elsevier B.V. All rights reserved.