Using computer simulations and high-resolution fMRI experiments in humans (n=6) and rats (n=8), we investigated to what extent BOLD fMRI temporal resolution is limited by dispersion in the venous vasculature. For this purpose, time-to-peak (TTP) and full-width at half-maximum (FWHM) of the BOLD impulse response (IR) function were determined. In fMRI experiments, a binary m-sequence probe method was used to obtain high-sensitivity model-free single-pixel estimates of IR. Simulations of postcapillary flow suggested that flow-related dispersion leads to a TTP and FWHM increase, which can amount to several seconds in larger pial veins. fMRI experiments showed substantial spatial variation in IR timing within human visual cortex, together with a correlation between TTP and FWHM. Averaged across the activated regions and across subjects, TTP and FWHM were 4.51+/-0.52 and 4.04+/-0.42 s, respectively. In regions of interest (ROI) weighted toward the larger venous structures, TTP and FWHM increased to 5.07+/-0.64 and 4.32+/-0.48 s, respectively. In rat somatosensory cortex, TTP and FWHM were substantially shorter than in humans (2.73+/-0.60 and 2.28+/-0.63 s, respectively). These results are consistent with a substantial macrovascular dispersive contribution to BOLD IR in humans, and furthermore suggest that neurovascular coupling is a relatively rapid process, with a resolution below 2.3 s FWHM.