We present different reaction models commonly used in nuclear astrophysics, in particular for the nucleosynthesis of the light elements. Nuclear reactions involved in stellar evolution generally occur at energies much lower than the Coulomb barrier. This property makes the cross sections extremely small, and virtually impossible to be measured in the laboratory. We start with a general discussion of low-energy scattering, and define the various cross sections required for reaction networks (essentially radiative capture and transfer reactions). Then we present specific models. Microscopic theories are based on fundamental principles, such as a nucleon-nucleon interaction, and an exact account of the antisymmetrization between all nucleons. In this context, most calculations performed so far have been done in the cluster approximation, but recent works, referred to as "ab initio" models, go beyond this approximation. Microscopic models can be simplified by neglecting the internal structure of the colliding nuclei, which leads to the potential model, also named the optical model. An alternative approach for the theoretical analysis of the experimental data is based on the phenomenological R-matrix theory, where parameters are fitted to the existing data, and then used to extrapolate the cross sections down to stellar energies. Indirect approaches, such as the Trojan Horse method, are briefly outlined. Finally, we present some typical applications of the different models.