We deployed a dense temporal seismic network in the source region of the 2004 mid-Niigata prefecture earthquake (thrust fault), Japan. A detailed velocity structure and accurate aftershock distributions were elucidated by inverting aftershock arrival times using double-difference tomography. A stress tensor inversion using the first-motion data was also conducted in order to investigate the stress field. The seismic velocities in the hanging wall above the main shock fault are lower than those in the footwall, with the velocity contrast extending to a depth of approximately 10 km. The aftershocks along the main shock rupture zone are distributed around the sharp boundary between the low- and high-velocity bodies. Furthermore, aftershocks associated with the largest aftershock appear to be aligned on a boundary between low- and high-velocity zones, in the footwall. The orientation of maximum principal stress (σ1) is consistent with the regional compressional strain rate axis inferred from GPS data, except in the southwestern side of the main shock hypocenter where the azimuth of σ1 rotates approximately 20¿ counterclockwise. The main shock hypocenter was located roughly at the transition zone where the structure of the hanging wall changes laterally and the azimuth of σ1 rotates. Heterogeneous structures of the seismic velocity and the stress field, combined with the ductile deformation of the upper crust, may have concentrated seismogenic stress around the hypocenter area to cause the complex distributions of aftershock sequence on structural boundaries.