The human colon is prone to develop a cancer due to its cell renovation that consists in a large number of cell divisions per day located in small cavities of the colon epithelium, called crypts. The colon epithelium is filled by millions of crypts, and it is known that mutations in the cell proliferation process (inside the crypts) can lead to the carcinogenesis. Colonic cell proliferation can be modeled by using multiscales (FIGUEIREDO et al., 2013). In particular, we can use a reference crypt, as a microscale domain, that is periodically distributed in a macroscale domain that is a portion of the colon epithelium. The final model results in a coupled system formed by an elliptic and parabolic equations whose unknowns are the proliferative cell density and the exerted cell pressure.We present a homogenization for the final PDE model where it is supposed to exist a asymptotic expansion for the exact solution of the problem , see (CIORANESCU; DONATO, 1999). We apply a multiscale method based on finite elements (HMM-FEM) to approximate the homogenized solution as in (ABDULLE, 2009; ABDULLE, 2012;ABDULLE; HUBER, 2014). The coupling and the non-linearity of the system implies a more complex implementation and increase the computational effort, thus we first solve the elliptic problem and then the parabolic one to make it easier. As we can see later, that strategy does not affect the convergence rates.Furthermore, in a single scale, we study a supraconvergent method based on centered finite difference to nonuniform mesh which is equivalent to a fully discrete linear finite element method. Firstly we study convergence and stability of a simpler model and then we prove for s " 1, 2 order Oph s q convergence of solution and gradient if the exact solution is in H 1`s pΩq, see BARBEIRO; GRIGORIEFF, 2005). Numerical results illustrate the methods above. For the multiscale problem, we present a supraconvergent scheme which provides approximations to the coupled system with quadratic convergence rate. This is done by solving the homogenized problem with the supraconvergent method discussed before. Our last contribution is a multiscale model in development which can be useful to solve multiscale problems with the good convergence rates discussed above. That model is based on solving a microscale problem that will be used to construct a macroscale solution for the homogenized system. Numerical results for this model suggest a supraconvergence.