In recent years, fiber reinforced cement-based materials have gained relevance in the civil engineering industry. Due to its excellent mechanical properties and contribution to crack propagation control, there is a great appeal to its usage as a construction material. However, technical standards for fiber reinforced concrete are still not established. A better understanding of the behavior of cement composite materials requires the representation of the material phases and their interfacial behavior. Stresses and strain distributions, damage evolution and fracture initiation develop at the observation scale of the heterogeneities and help to explain and predict the behavior of concrete at a macroscopic level. The numerical modeling of these composites emerge as challenging and complex problems. For this, it is necessary to define the main mechanisms that describe the material behavior in order to choose the proper mathematical formulation. This dissertation proposes methodologies for the numerical modeling of cement composite materials in a multiscale approach. From the information obtained at the material scale, this work aims at assessing the global behavior of the composite. Numerical and computational procedures will be developed based on the Finite Element Method, Artificial Intelligence techniques and concepts of Computational Damage Mechanics. At the macroscale, an equivalent continuum model is developed through probabilistic and Artificial Intelligence techniques. At the mesoscale, two approaches are proposed. The first includes the fibers through interface elements. The second adopts a new fiber-matrix composite element. With the models developed here, it is possible to evaluate damage evolution, fracture propagation patterns, load-displacement global behavior of the composite upto failure. Experimental results from the literature give support to the conclusions.