Rutger,
I'm glad Doug replied to your question, since he's the expert, and he also gives more thorough explanations than I do.
I was also going to suggest using 3dDeconvolve to subtract the baseline condition from the test condition within each subject before doing the group analysis.
One thing Doug didn't mention is the necessity of co-registering data for the two sessions of each subject. We do it by using the 3dvolreg program in AFNI to align the high resolution anatomical scans from the two sessions, and then applying this rotation/translation to the functional data. The details on this can be found in Bob Cox's document README.registration: afni.nimh.nih.gov/afni/docREADME/README.registration
Sally