We use the SDSS/DR8 galaxy sample to study the radial distribution of satellite galaxies around isolated primaries, comparing to semi-analytic models of galaxy formation based on the Millennium and Millennium-II simulations. SDSS satellites behave differently around high- and low-mass primaries: those orbiting objects with $M_*>10^{11}M_odot$ are mostly red and are less concentrated towards their host than the inferred dark matter halo, an effect that is very pronounced for the few blue satellites. On the other hand, less massive primaries have steeper satellite profiles that agree quite well with the expected dark matter distribution and are dominated by blue satellites, even in the inner regions where strong environmental effects are expected. In fact, such effects appear to be strong only for primaries with $M_* > 10^{11}M_odot$. This behaviour is not reproduced by current semi-analytic simulations, where satellite profiles always parallel those of the dark matter and satellite populations are predominantly red for primaries of all masses. The disagreement with SDSS suggests that environmental effects are too efficient in the models. Modifying the treatment of environmental and star formation processes can substantially increase the fraction of blue satellites, but their radial distribution remains significantly shallower than observed. It seems that most satellites of low-mass primaries can continue to form stars even after orbiting within their joint halo for 5 Gyr or more.