The paper presents an improved grid-based Bader analysis algorithm that efficiently and robustly partitions a charge density grid into Bader volumes, maintaining linear scaling with the number of grid points. The algorithm follows steepest ascent paths along the charge density gradient from grid point to grid point until a charge density maximum is reached. The key innovation is the representation of accurate off-lattice ascent paths with respect to the grid points, eliminating the lattice bias that previously caused Bader surfaces to align along grid directions. This modification retains the efficiency and linear scaling of the original algorithm while ensuring more accurate and unbiased results. The method is particularly useful for large systems and complex bonding geometries, as demonstrated through various tests, including a two-dimensional model, a water molecule, and an ionic charge calculation in a NaCl crystal. The near-grid method shows monotonic and smooth convergence with increasing grid density, outperforming the on-grid method in terms of accuracy and systematic convergence.The paper presents an improved grid-based Bader analysis algorithm that efficiently and robustly partitions a charge density grid into Bader volumes, maintaining linear scaling with the number of grid points. The algorithm follows steepest ascent paths along the charge density gradient from grid point to grid point until a charge density maximum is reached. The key innovation is the representation of accurate off-lattice ascent paths with respect to the grid points, eliminating the lattice bias that previously caused Bader surfaces to align along grid directions. This modification retains the efficiency and linear scaling of the original algorithm while ensuring more accurate and unbiased results. The method is particularly useful for large systems and complex bonding geometries, as demonstrated through various tests, including a two-dimensional model, a water molecule, and an ionic charge calculation in a NaCl crystal. The near-grid method shows monotonic and smooth convergence with increasing grid density, outperforming the on-grid method in terms of accuracy and systematic convergence.