Particulate composites are composed of random packing of polydispersed particles, such as cementitious composites. The packing behavior of particles plays an important role in the mesostructural evolution of particulate composites. The preponderance of previous works has focused on the random packing of spherical, cylindrical or elliptical particles, and little is known about ellipsoidal particles. In the present work, a contact detection algorithm of ellipsoids is presented in detail, which can quantitatively describe relative positions of ellipsoids in a three-dimensional space. The accuracy and efficiency of the algorithm are tested and compared with algorithms developed in previous literature, and its reliability is verified by a random sequential packing model of polydispersed ellipsoids with semi-periodic boundary conditions. Moreover, the algorithm is applied to simulate a mesostructure model of cementitious composites. By means of stereological tools and section analysis technologies, the mesostructure of cementitious composites is quantitatively characterized. Finally, the reliability of the statistical results from the present methodology is verified by valid experimental and theoretical results. (c) 2012 Elsevier B.V. All rights reserved.