For the purposes of your project, it is perfectly fine to present your results on a subject-by-subject basis. Therefore, you don't really need to do a 3rd-level (i.e., cross-subject level analysis). Nevertheless, some of you may want to try it (and have asked how to do it) and, frankly, it may bring out effects you weren't able to see at the subject level. The main thing to keep in mind is that your 3rd level analysis should be a fixed-effects analysis. I suspect you will all be able to justify this if you choose to do the 3rd level.
The first thing to keep in mind about you third level analysis is that you will do one for every COPE (or contrast) you had at the lower level. So, if you had 6 contrasts at the first level, you should have 6 cope#.feat directories in your second level *.gfeat directory. If you want to see the group level effects for all of these COPEs, then you will do 6 third level analyses, and each one will use the cope image inside the stats directory of these cope#.feat directories (see below).
Once you open the FEAT GUI, the first thing you'll need to do is change the analysis type (in the upper left corner) to “Higher-level Analysis”, just as if you were doing a second level analysis.
Ignore the paths below. I edited a version I found for this description and I did not change the paths since they do not matter anyway.
/usr/local/sharity/var/mount/Goldman.Data/BIAC/Percept.02/Data/FSL/32894/2ndLevel.gfeat/cope1.feat/stats/cope1.nii.gz
/usr/local/sharity/var/mount/Goldman.Data/BIAC/Percept.02/Data/FSL/32845/run01.feat/stats/cope1.nii.gz
Now click “Go” and it should work. Repeat for your other COPES.