afni.nimh.nih.gov

Well, you could model the responses for each task separately if you expect them to be different. If you have plenty of data points relative to the number of regressors you are using, then using separate responses would not be a costly issue.
You can always try it both ways and see what you get. I expect, little if any difference in the regressors of interest's coefficients.

cheers,
ziad