Three methods of variable-weight statistical enhancement for Monte Carlo semiconductor device simulation are compared. The steady-state statistical errors and figures of merit for implementations of the multicomb, cloning-rouletting, and splitting-gathering enhancement methods are obtained for bulk silicon simulations. The results indicate that all methods enhance the high-energy distribution tail with comparable accuracy, but that the splitting-gathering method achieves a lower error at low energies by automatically preserving a peak in the bin populations at the peak of the particle energy distribution.