The only time normalization will "foul" your sound is when you use the RMS option. That mode actually applies compression to the sound on a dynamic basis, which can sound good, but not always.
Otherwise (using peak level) it simply scales the waveform so that the highest peak moves to 0dB (or whatever you set it to). It essentially does EXACTLY what DarkstarJedi says to do, but automated.
The only time I ever use RMS normalize is when I've got a so-so recording that I otherwise would not listen to, save for the fact that it's my only copy.