Due to the relatively complex nature of spectral measurements from x-ray machines, many researchers use mathematical models to simulate the spectra they need. However, there is concern over their accuracy, and hence the impact that their accuracy may have, on subsequent calculations that rely upon the spectra modelled. With this in mind spectral measurements have been performed on a mammography machine and a comparison with spectra calculated using several different models is presented. Several different techniques have been investigated in the spectral measurements to allow for pulse pileup and other effects of high count rate. Comparison with half value layer (HVL) measurements shows that the use of a gating signal in conjunction with the air-free path provides accurate results without the need for a pinhole collimator. Comparison of the measured spectra with those calculated using different models proposed in the literature suggests that accurate results can be produced by all models, but only if the user attempts to match the calculated HVL of the modelled spectrum with the physically measured HVL. If this is not done the modelled spectra may be in error. The impact of such an error is demonstrated in calculations of mean glandular dose, which indicate a possible underestimate of the dose by up to 20%.