When multiple alignment is performed between a number of related sequences, in some cases the sequences may have different lengths. What should I do with these unaligned fragments?
If you have short or long insertion in one of your sequences (but not in the others), it´s uninformative for phylogenetic analyses (informative for species/population identification only) as such - you can delete it from your alignment by hand or use Gblocks as suggested above.
We spent quite a bit of time discussing homologous and non-homologous sequences in a recent paper (open access): http://mbe.oxfordjournals.org/content/31/3/501