I'd be interested in hearing a definitive answer to this, also. My recollection from years ago was that Doug Ward was pretty adamant about running AlphaSim in original space, in which case one would need to multiply by the original voxel dimensions to use the results to threshold in 1x1x1 Talairach space. In more recent message board posts, the recommendations range from running AlphaSim in Talairach space to begin with (for group data) to resampling to 2x2x2 and splitting the difference (more or less). So I'm a little confused about the departure from what was originally recommended by the person who wrote the program. Have we since learned that the difference in results is trivial enough that it doesn't really matter?