We discuss the uncertainties and the systematic effects that exist in the estimates of the stellar masses of high redshift galaxies, using broad band photometry, and how they affect the deduced galaxy stellar mass function. We use at this purpose the latest version of the GOODS-MUSIC catalog. In particular, we discuss the impact of different synthetic models, of the assumed initial mass function and of the selection band. Using Charlot & Bruzual 2007 and Maraston 2005 models we find masses lower than those obtained from Bruzual & Charlot 2003 models. In addition, we find a slight trend as a function of the mass itself comparing these two mass determinations with that from Bruzual & Charlot 2003 models. As consequence, the derived galaxy stellar mass functions show diverse shapes, and their slope depends on the assumed models. Despite these differences, the overall results and scenario remains unchanged. The masses obtained with the assumption of the Chabrier initial mass function are in average 0.24 dex lower than those from the Salpeter assumption, at all redshifts, causing a shift of galaxy stellar mass function of the same amount. Finally, using a 4.5 um-selected sample instead of a Ks-selected one, we add a new population of highly absorbed, dusty galaxies at z≃ 2-3 of relatively low masses, yielding stronger constraints on the slope of the galaxy stellar mass function at lower masses. ; Comment: 5 pages, 4 figures, proceedings of the conference "Probing Stellar Populations out to the Distant Universe", Cefalu (Italy), September 7 - 19, 2008. To be published in the AIP Conf. Proc. Series