Abstract: The accurate estimation of system harmonic impedance is critical to realize the quantitative determination of harmonic responsibility. The customer-side harmonic impedance is no longer much greater than that of the system side in the situation of new energy resources connecting to the grid, which results in accuracy decrease or ineffectiveness of existing estimation methods. This paper proposes an estimation method of system harmonic impedance based on the sub-space decomposition and the dynamic coefficient regression. The observed signals of the harmonic voltage and current at the point of common coupling（PCC） are decomposed into several sub-spaces by the wavelet packet decomposition. The sub-space with the weakest correlation between the explanatory variables is selected according to the mutual information value, which reduces the impact of the correlation between explanatory variables on the regression analysis. Considering that the system harmonic fluctuation will interfere with the correlation between the harmonic voltage and current at PCC, the system harmonic voltage is regarded as a dynamic coefficient. The system harmonic impedance is calculated by the dynamic coefficient regression method, to reduce the impact of harmonic voltage fluctuation on the estimation results. The simulation results show that the proposed method has better estimation accuracy and robustness compared with the existing methods.