Gap penalties are a field where a great deal more work is required. They are often applied without much justification. Dayhoff suggested using a gap penalty of 6 with PAM250 matrices. Henikoff & Henikoff suggest using a gap penalty of 8 with BLOSUM62 matrices. Both have little reason, other than empirical support, for their choices. The MEGA program package and the PHYLIP program package go to the extreme of ignoring all gaps and any missing data. They do this because there is no accurate way to weight changes due to indels releative to substitutions. Never-the-less, the indels do contain some information but the current challenge is to correctly extract it in a precise manner. In this respect, the approach being taken by Thorne (mentioned in the section on alignments) holds great promise. Barring such a sophisticated approach, it is suggested that you use a variety of gap penalties (from zero to some significant punishment) and from these determine the effects that this has on your results.