Time-lapse seismology is important for monitoring subsurface pressure changes and fluid movements in producing hydrocarbon reservoirs. We analyse two 4-D, 3C onshore surveys from Vacuum Field, New Mexico, USA, where the reservoir of interest is a fractured dolomite. In Phase VI, a time-lapse survey was acquired before and after a pilot tertiary-recovery programme of overpressured CO2 injection, which altered the fluid composition and the pore-fluid pressure. Phase VII was a similar time-lapse survey in the same location but with a different lower-pressure injection regime. Applying a processing sequence to the Phase VI data preserving normal-incidence shear-wave anisotropy (time-delays and polarization) and maximizing repeatability, interval-time analysis of the reservoir interval shows a significant 10 per cent change in shear-wave velocity anisotropy and 3 per cent decrease in the P-wave interval velocities. A 1-D model incorporating both saturation and pressure changes is matched to the data. The saturation changes have little effect on the seismic velocities. There are two main causes of the time-lapse changes. Any change in pore-fluid pressures modifies crack aspect ratios. Additionally, when there are overpressures, as there are in Phase VI, there is a 90degrees change in maximum impedance directions, and the leading faster split shear wave, instead of being parallel to the crack face as it is for low pore-fluid pressures, becomes orthogonal to the crack face. The anisotropic poro-elasticity (APE) model of the evolution of microcracked rock, calculates the evolution of cracked rock to changing conditions. APE modelling shows that at high overburden pressures only nearly vertical cracks, to which normal incidence P waves are less sensitive than S waves, remain open as the pore-fluid pressure increases. APE modelling matches the observed time-lapse effects almost exactly demonstrating that shear-wave anisotropy is a highly sensitive diagnostic of pore-fluid pressure changes in fractured reservoirs. In this comparatively limited analysis, APE modelling of fluid-injection at known pressure correctly predicted the changes in seismic response, particularly the shear-wave splitting, induced by the high-pressure CO2 injection. In the Phase VII survey, APE modelling also successfully predicted the response to the lower-pressure injection using the same Phase VI model of the cracked reservoir. The underlying reason for this remarkable predictability of fluid-saturated reservoir rocks is the critical nature and high crack density of the fluid-saturated cracks and microcracks in the reservoir rock, which makes cracked reservoirs critical systems.