> When I use the wild card in 3dDeconvolve, Isn't each run modeled separately with its
> own set of regressors for that run?

No, there is only one model for all the data. However, there are some subtleties. For example,

1) discontinuities are allowed for slow drifts across runs;
2) discontinuities across runs are probably better modeled for motion effects.

> If thats not the case, Using 3dDeconvolve Do I have to do the regression separately for each run?

What I meant in the first suggestion of my previous response is, instead of using 6 motion regressors in the model, try feed in 3dDeconvolve with 6*n motion regressors, where n is the number of runs. If you use uber_subject.py or afni_proc.py, there is option that automatically creates 6 motion regressors for each run separately.

Gang