A generalised Voronoi tessellation is proposed to create three-dimensional microstructural finite element model, which can effectively reproduce the grain size distribution and grain aspect ratio obtained from experiments. This new approach consists of two steps. The first step generates the desired lognormal grain size distribution with a given average grain volume and standard deviation. The second step requires grouping meshed elements to create a specific grain aspect ratio, using the Voronoi generators from the first step. A new concept is introduced to describe the transition from the Poisson-Voronoi tessellation to the centroidal Voronoi tessellation. More importantly, instead of using the conventional way where the Voronoi cells are first generated and then meshed into finite elements, this new approach discretises the pre-meshed specimen with the Voronoi generators. This new technique prevents the presence of high density mesh at the vertices of Voronoi cells, and can tessellate irregular geometry much more easily. Examples of microstructures with different size distributions, non-equiaxed grains and complicated specimen geometries further demonstrate that the proposed approach can offer great flexibility to model various specimen geometries while keeping the process simple and efficient. GENERALISED VORONOI TESSELLATION FOR GENERATING MICROSTRUCTURES 145 polycrystals. Its effectiveness and robustness have been verified for CPFE applications in terms of nanoindentation, fatigue, grain size and grain boundary effects and twinning [7][8][9][10][11]. Kumar et al. [12] studied the properties of 3D Poisson-Voronoi cells in terms of the distributions of the number of cell faces, volume and surface area. They numerically simulated several hundred thousand cells using the Monte Carlo method and found that the faces, volume and surface area distributions of the Poisson-Voronoi cells can be best described by a two-parameter gamma distribution. Nonetheless, if the number of cells is less than several thousands, a lognormal distribution can also be used to approximate these distributions. In terms of grain volume, intersected area and intercept length in a polycrystal, Vaz and Fortes [13] concluded that these properties are usually lognormally distributed. Their experimental results have shown that the gamma and lognormal distributions can describe the sizes of metallic polycrystals fairly well. Furthermore, it was revealed that the distributions of different dimensions, that is cell volumes in 3D, cell areas in 2D and edge lengths in 1D, have to be fitted with different parameters of distribution [14]. Hence, in order to accurately study a 3D microstructure involving the grain size distribution using CPFE simulations, a 3D VT that reflects the specific grain size distribution is required. The lognormal distribution function was also used to study elastic-viscoplastic materials through a representative volume element [15]. Recently, Lazar et al. [16] analysed a structure containing 250 million Poisson-Voronoi cell...