For "fun" I decided to add spm and afni to my set of voxel indexes started with MRIcroN, R, and pyMVPA.
SPM8 follows the 1-based style of MRIcroN and R:
while afni follows the 0-based style of pyMVPA:
So, it's probably not fair to think of either the one or zero-based styles as "correct", but this needs to be kept in mind when referring to voxel by i,j,k coordinates programmatically.
I don't use fsl or brain voyager; send me how these (or other) programs show the voxel coordinates and I'll add them to the collection.
Update 9 August 2013: fsl and caret are both 0-based.