We analysed red giant branch stars in 16 Galactic globular clusters , computing their atmospheric parameters both from the photometry and from excitation and ionisation balances .
The spectroscopic parameters are lower than the photometric ones and this discrepancy increases decreasing the metallicity , reaching , at [ Fe/H ] \sim –2.5 dex , differences of \sim 350 K in effective temperature and \sim 1 dex in surface gravity .
We demonstrate that the spectroscopic parameters are inconsistent with the position of the stars in the colour-magnitude diagram , providing too low temperatures and gravities , and predicting that the stars are up to about 2.5 magnitudes brighter than the observed magnitudes .

The parameter discrepancy is likely due to the inadequacies of the adopted physics , in particular the assumption of 1-dimensional geometry can be the origin of the observed slope between iron abundances and excitation potential that leads to low temperatures .
However , the current modelling of 3D/NLTE radiative transfer for giant stars seems to be not able to totally erase this slope .

We conclude that the spectroscopic parameters are wrong for metallicity lower than –1.5 dex and for these red giant stars photometric temperatures and gravities should be adopted .
We provide a simple relation to correct the spectroscopic temperatures in order to put them onto a photometric scale .