The stellar population models dramatically progressed with the arrival of large and complete libraries, ELODIE, CFLIB (=Indo-US) and MILES at a relatively high resolution. We show that the quality of the fits is not anymore limited by the size of the stellar libraries in a large range of ages (0.1 to 10 Gyrs) and metallicities (-2 to +0.4 dex). The main limitations of the empirical stellar libraries are (i) the coverage of the parameters space (lack of hot stars of low metallicity), (ii) the precision and homogeneity of the atmospheric parameters and (iii) the non-resolution of individual element abundances (in particular [$\alpha$/Fe]). Detailed abundance measurements in the large libraries, and usage of theoretical libraries are probably the next steps, and we show that a combination between an empirical (ELODIE) and a theoretical library (Coelho et al. 2005) immediately improves the modeling of ($\alpha$-enhanced) globular clusters.