[Federico]

I usually have to normalize some tracks when mixing music I didn't record.
I use this rules of thumb:

First: I normalize only when I can gain a fair auount of dBs (8dB for me).
I try not to look at peaks but at RMS. For example I remember a vocal track
(I did not recorded it) where 99% of the time the level was around -18dB and
in one point the singer yelled at -1dB.
You don't want to normalize a track like that because you can only gain
0.8dB before digital clipping. In that case I sliced off the yell and then
normalized the rest of the track.

Second: normalize but keep a "headroom". If I am in a need to normalize I do
it to -2dB. So you can put a digital eq. and you can boost frequencies
without reducing the eq input gain.

Third: I never normalize only one track of a stereo pair. It depends on the
software but I do not want to mess up the stereo image playing with phases.
Or at least it must be a choice of mine!

F.