This paper presents the solutions obtained with the coarse mesh transport code COMET to the 3-D C5G7 MOX benchmark problem. The coarse mesh transport method uses precomputed heterogeneous response functions for unique coarse meshes in the core to generate the detailed whole core solution such as the fuel pin power distribution. This is achieved by converging on the coarse mesh interface incident fluxes that are approximated by truncated Legendre polynomials in angular and spatial variables. The solutions to all three configurations of the benchmark problem are obtained by using 2nd order expansions in all variables on the interface. The precomputed response functions, generated (tallied) by Monte Carlo means (a modified version of MCNP4C), are responses to a set of coarse mesh interface incident fluxes given in terms of the normalize Legendre polynomials in polar, azimuthal and spatial variables. As compared to the MCNP reference solutions, the unrodded and rodded A and rodded B cases have an eigenvalue relative error of 0.06% with a standard deviation of 0.02%. The average relative errors for the pin power in all configurations are 1.0%, 1.0% and 1.1%, respectively. The maximum errors are all located in the low power region of axial slice 3 with values 10.3%, 10.9% and 11.2% for the unrodded, rodded A and B configurations, respectively.