Diffusion fMRI (dfMRI) has been proposed as a more direct means for mapping neural activity more accurately than BOLD fMRI. However, the origin of dfMRI signals is still an ongoing debate. Here, we developed a line-scanning dfMRI technique achieving very high temporal resolution (100 ms), and measured activity in the forelimb S1 upon rat forepaw stimulation. Our results show a rapid-onset (<200ms) dfMRI component that was not found in BOLD fMRI. Upon inducing hypercapnia, the fast dfMRI component was nearly unaffected while the slower dfMRI component was substantially modulated, suggesting a potentially neural origin for the former.