The estimation of the mean glandular dose in mammography using Monte Carlo simulations requires the calculation of the incident air kerma evaluated on the breast surface. In such a calculation, caution should be applied in considering explicitly the presence of the top compression paddle, since Compton scattering in this slab may produce a large spread of the incidence angles of x-ray photons on the scoring surface. Then, the calculation of the incident air kerma should contain the 'effective' area of the scoring surface, which takes into account the angle of incidence of photons on such a surface. Using Geant4 Monte Carlo simulations with a code previously validated according to the Task Group 195 of the American Association of Physicists in Medicine, we show that for typical x-ray spectra and energy range adopted in mammography, the resulting discrepancy in the calculation of the incident air kerma may lead to an overestimation from a minimum of 10% up to 12% of normalized dose coefficients and, hence, of the corresponding mean glandular dose if this contribution is not considered.