Hi Luis,

Could I suggest that you read this article,
  http://dx.doi.org/10.1016/j.neuroimage.2...
and decide whether you really want to "Talairach" the data (i.e. determine a piece-wise linear map, with the aim of comparing to the published atlas) or to determine a more flexible nonlinear warp to a more representative template?

Most commonly used neuroimaging software packages have a way of spatially normalising data to MNI/ICBM space. Depending on the data and the goals, you might prefer to do group-wise registration of your subjects to their own mean (which can be more precise and less biased) and then to map this mean to ICBM space. For example, see page 444 (and also the rest of this fairly short chapter) of the SPM manual for instructions on how to achieve this using DARTEL:
http://www.fil.ion.ucl.ac.uk/spm/doc/man...

Note that you can warp to an ICBM/MNI template and then derive Talairach coordinates if necessary, see e.g.
  http://www.brainmap.org/icbm2tal/

I hope that helps,
Ged