Dummy feature filling is an efficient approach for reducing wafer-topography variation in chemical-mechanical polishing (CMP), which is the key planarization process in modern VLSI fabrication. In this paper, we present a new min-variance iterative method for fast smart dummy feature density assignment and post-CMP topography variation reduction. This method iteratively selects target areas using an efficient CMP low-pass filter model and a variance-minimizing heuristic, and assigns/removes dummy features accordingly. Because of the efficient usage of the 2-D Fast Fourier Transform (FFT), the computational cost of this new method is close to O(nlog(n)), making it much faster than the existing linear programming method that costs O(n^3 ). Numerical experiments show the computational cost of our new method is almost negligible when compared with the LP method and its solution is very close to the optimal solution.