Jane,

Sorry that I did not quite understand your concern the first time.

For consistency normalization is usually done after registration, but I don't see such a normalization order as a problem. For example, if you have four runs during the 2nd day with each run having 92 time points, you can get their means first:

3dTstat -mean -prefix mean1 dataset+orig'[<0>..<91>]'
3dTstat -mean -prefix mean2 dataset+orig'[<92>..<183>]'
3dTstat -mean -prefix mean3 dataset+orig'[<184>..<275>]'
3dTstat -mean -prefix mean4 dataset+orig'[<276>..<367>]'

Then normalize each run:

3dcalc \
-a dataset+orig'[<0>..<91>]' \
-b mean1+orig \
-expr "a/b*100" \
-prefix dataset.norm1

3dcalc \
-a dataset+orig'[<92>..<183>]' \
-b mean2+orig \
-expr "a/b*100" \
-prefix dataset.norm2

3dcalc \
-a dataset+orig'[<184>..<275>]' \
-b mean3+orig \
-expr "a/b*100" \
-prefix dataset.norm3

3dcalc \
-a dataset+orig'[<276>..<367>]' \
-b mean4+orig \
-expr "a/b*100" \
-prefix dataset.norm4

Concatenate the four normalized runs:

3dTcat -prefix dataset.norm \
dataset.norm1+orig \
dataset.norm2+orig \
dataset.norm3+orig \
dataset.norm4+orig

Gang