The use of model atmospheres for deriving stellar fundamental parameters, such as Teff, log(g), and [Fe/H], will increase as we find and explore extreme stellar populations where empirical calibrations are not yet available. Moreover, calibrations for upcoming large satellite missions of new spectrophotometric indices, similar to the uvby-Hbeta system, will be needed. We aim to test the power of theoretical calibrations based on a new generation of MARCS models by comparisons with observational photometric data. We calculated synthetic uvby-Hbeta colour indices from synthetic spectra. A sample of 367 field stars, as well as stars in globular clusters, is used for a direct comparison of the synthetic indices versus empirical data and for scrutizing the possibilities of theoretical calibrations for temperature, metallicity, and gravity. We show that the temperature sensitivity of the synthetic (b-y) colour is very close to its empirical counterpart, whereas the temperature scale based upon Hbeta shows a slight offset. The theoretical metallicity sensitivity of the m1 index (and for G-type stars its combination with c1) is somewhat higher than the empirical one, based upon spectroscopic determinations. The gravity sensitivity of the synthetic c1 index shows satisfactory behaviour when compared to observations of F stars. For stars cooler than the sun, a deviation is significant in the c1-(b-y) diagram. The theoretical calibrations of (b-y), (v-y), and c1 seem to work well for Pop II stars and lead to effective temperatures for globular cluster stars supporting recent claims that atomic diffusion occurs in stars near the turnoff point of NGC 6397. Synthetic colours of stellar atmospheres can indeed be used, in many cases, to derive reliable fundamental stellar parameters. The deviations seen when compared to observational data could be due to incomplete linelists but are possibly also due to the effects of assuming plane-parallel or spherical geometry and LTE.