Hi Adam,
If you wish to do an ANCOVA, you might find
3dttest++ or
3dMVM easier.
If you are absolutely determined to use 3dRegAna, you will need to "dummy code" your task variable. You can see Example 4 in the
3dRegAna white paper for guidance. If you have repeated-measures (i.e. everyone in the experiment did all 4 tasks) then things get even more interesting...
It's also possible that I missed your question, so feel free to clarify if so.
Peter