> I guess the other way to approach this would be to concatenate all 7 runs,
> concatenate the stim_files, and concatenate the stim_times. However, I
> choose to do each run separately since it would take some time to reorganize
> my stim_times files to be global since they are currently in local format.

There is an option in 3dDeconvolve, called -local_times, with which you can specify each run's stimuli in local times.

> Then I use 3dcalc to take the mean of the 7 condition_B_correlation t-maps
> and the mean of the 7 condition_A_correlation t-maps. Once I have these for
> all my subjects I can t-test them with 3dtest.

By t-maps do you mean that you average the t-statistic or beta value across runs? It should be the beta values that get averaged. However if you want to take the correlation coefficients for group analysis, it's not a good idea to average them either.

> Although this may not be the most efficient way to do this, is this still correct?

Yes, it should be more or less close to the alternative approach.

Gang