I build two models for the same subject's session: one for normalized data (2x2x2 voxel size) and once for non-normalized data (1.56x1.56x2.4 voxel size). The data matrix in first case is 79x95x69 and in the second 128x128x24. The volume of one voxel in first case is much larger (8 mm^3 vs. 5.84 mm^3). 

My puzzle is that the total number of brain voxels in a mask.img is more or less the same in both cases. Obviously, sizes of activations in voxels number is also  more or less the same. It looks like the resolution for non-normalized data is rounded somehow. Does it mean that I cannot trust the volume size of non-normlaized data? I am checked using SPM5.

Thanks for help,

Vadim