thanks for the magnanimous words of encouragement (and subtle mathematical reprimenda...point taken!). I did do some comparisons using 3dDeconvolve on a single-voxel timeseries and the statistics improved with the dct-baseline model (lower p-values for both the stimuli of interests and for the full model). And, yes, I found the 1deval *very* useful in constructing the dct set... every time I need to do some custom data manipulation I cannot but marvel at the farsightedness of Zorkon's mind in laying down all the necessary bricks...
devotely
giuseppe