Grids of stellar spectra are extensively used in astrophysics in a range of applications, from derivation of atmospheric parameters in observed stellar spectra, to archaeological studies in integrated light of galaxies. Theoretical stellar libraries have been increasingly used in recent years to overcome limitations of empirical libraries by, for example, exploring atmospheric parameter spaces not well represented in the latter. Nevertheless, there are still caveats and stellar libraries can only partially model observations with accuracy. Here we discuss the current need for synthetic libraries, in particular for applications in stellar population studies. Next we briefly present a non-extensive sample of libraries available in the literature, and we then highlight how better atomic and molecular opacities are much needed to overcome the deficiencies of current synthetic libraries.