We present a detailed analysis for the effect of stellar feedback on both the
stellar population of 45 individual galaxies using a sample of high-resolution
cosmological zoom simulations (10^10<M_halp<10^13 Msun) and on full galaxy
populations using newly developed semi-analytic models applied to merger trees
of the Millennium simulation.
Regarding the zoom simulations which include an empirical model for
momentum-driven stellar winds, stellar winds are found to be successful in
suppressing early star formation, in increasing cold gas fractions (due to
re-accretion of gas) and thus, in reducing the baryon conversion efficiencies
for low mass halos (<10^12 Msun) in reasonable agreement with observations at
various cosmic epochs from z=3 to the present day. Furthermore, metal cooling
and stellar winds are both found to significantly increase the ratio of in-situ
formed to accreted stars by a factor of more than 10 roughly consistent with
predictions from recent sub-halo abundance matching (SHAM) models. To
additionally test the effect of stellar feedback on statistical properties of
full galaxy populations, in particular on the stellar mass function, we take
advantage of semi-analytic models including different models for the
self-regulation between star formation, stellar feedback and re-accretion of 
gas (partly based on the zoom simulation results). Our preliminary results 
indicate that the number density of low-mass galaxies is particularly 
sensitive to the amount of gas which gets entirely ejected out of the halo 
(and is not only heated). Interestingly, and in contrast to previous studies, 
we cannot confirm that only changing the re-incorporation time-scales is 
sufficient to bring the low-mass end of the stellar mass function at high 
redshifts in agreement with observations.