Currently, commodity graphics hardware have been increasingly used for solid texture based volume rendering. Since solid texture accepts only voxel dataset, voxelization techniques become indispensable for generating volume rendering images from irregular grids such as tetrahedral or hexahedral grids. In this paper, we propose a technique for voxelizing hexahedral grids by projecting every hexahedra onto a projection plane and scan-converting between the paired cell faces along the depth direction. Conventional voxelization techniques first convert hexahedral grids into tetrahedral grids by decomposing the hexahedral grid cells into five tetrahedral grid cells each, and then generate voxels by voxelizing these tetrahedral grids. This tetrahedralization leads to a decrease in accuracy of generated voxels because it considers only the four points of a tetrahedal cell during the data interpolation. We calculated the error generated from the tetrahedralization of a single hexahedral cell, and visualized the error distribution. Voxelization techniques for tetrahedral cells calculate in single pass the numerical data values for each voxel position of interporlating those defined at vertices of the tetrahedral grid cell. Our technique first interpolates local coordinates from the coordinates defined at the vertices of the hexahedral grid cell, and then calculates numerical data values using interpolated coordinates at voxel positions. This is the reason that we call "two-pass rasterization" technique. We used a set of hexahedral grid data for performance evaluation and compared the resulting data values with those obtained using conventional method. For this comparison we could confirm the effectiveness of the proposed technique.