In conclusion, we have presented a new and easy to implement method for calculating kinetic energy matrix elements of localised functions represented on a regular real space grid. This FFT box method is based on a mixed real space - reciprocal space approach. We use well established FFT algorithms to calculate the action of the Laplacian operator on localised support functions, whilst maintaining linear-scaling with system size and near locality of the operation. This makes our FFT box method suitable for implementation in the order-N code that we are developing. We have performed tests of the FFT box method and various orders of FD. Comparing to the exact integrals of the continuous representation, we have demonstrated that our approach is more accurate than low order FD approximations and only when A = 28 FD is used does the accuracy become comparable to that of the FFT box method. We have also highlighted the connection between the FFT box method and plane-wave methods and shown that our approach is up to three orders of magnitude more accurate than A = 28 FD when compared to the `exact' result within the plane-wave basis set of the entire simulation cell. Furthermore, our approach for calculating the kinetic energy is consistent with the way in which other quantities in a total-energy calculation, such as the electron density and the Hartree energy, are computed as these are also calculated using reciprocal space techniques. Finally, we also note that our FFT box method is more versatile than FD as it is applicable to real space grids based on any lattice symmetry whereas FD schemes are usually only applied to orthorhombic grids.