We present a high-resolution three-dimensional (3-D) anisotropic P wave velocity (Vp) model in the northern Hikurangi margin offshore Gisborne, New Zealand, constructed by tomographic inversion of over 430,000 first arrivals recorded by a dense grid of ocean bottom seismometers. Since the study area covers a region where shallow slow slip events (SSEs) occur repeatedly and the subduction of a seamount is proposed, it offers an ideal location to link our understanding of structural and hydrogeologic properties at megathrust faults to slip behavior. The Vp model reveals an~30-km-wide, low-velocity accretionary wedge at the toe of the overriding plate, where previous seismic reflection studies show a series of active thrust faults branching from the plate interface. We find some locations with significant Vp azimuthal anisotropy >5% near the branching faults and the deformation front. This finding suggests that the anisotropy is not ubiquitous and homogeneous within the overriding plate, but more localized in the vicinity of active thrust faults. The fast axes of Vp within the accretionary wedge are mostly oriented to the plate convergence direction, which is interpreted as preferentially oriented cracks in a compressional stress regime associated with plate subduction. We find that the magnitudes of anisotropy are roughly equivalent to values found at oceanic spreading centers, where the extensional stress regime is dominant and the crack density is expected to be higher than subduction zones. This consideration may indicate that additional effects such as fault foliation and clay mineral alignment also contribute to upper plate anisotropy along subduction margins. Plain Language Summary We use man-made sound waves (controlled-source seismic data) to reveal in three dimensions the speed of sound within rocks and sediments that make up the northern Hikurangi margin off the east coast of North Island, New Zealand. We find that the sediments lying on top of subducting plate (accretionary wedge) have low velocities and host several branching faults. We also find that the speed of sound is faster when measured in the direction that the plates are colliding and slower when measured parallel to the faults (this phenomenon is called seismic anisotropy). This difference in speed suggests that the faults contain cracks that are aligned parallel to the direction of the plate subduction.