Self-healing cementitious materials with microcapsules are complex multiscale and multiphase materials. The random microstructure of these materials governs their mechanical and transport behaviour. The actual microstructure can be represented accurately with a discrete lattice model, but computational restrictions mean that the size of domain that can be considered with this approach is limited. By contrast, a smeared approach, based on a micromechanical formulation, provides an approximate representation of the material microstructure with low computational costs. The aim of this paper is to compare simulations of a microcapsule-based self-healing cementitious system with discrete-lattice and smeared-micromechanical models, and to assess the relative strengths and weaknesses of these models for simulating distributed fracture and healing in this type of self-healing material. A novel random field generation technique is used to represent the microstructure of a cementitious mortar specimen. The meshes and elements are created by the triangulation method and used to determine the input required for the lattice model. The paper also describes the enhancement of the TUDelft lattice model to include self-healing behaviour. The extended micromechanical model considers both microcracking and healing. The findings from the study provide insight into the relative merits of these two modelling approaches.