Accurate estimation of system harmonic impedance is of great significance for harmonic responsibility division and harmonic traceability. It is proposed to use the Joint Approximate Diagonalization of Eigen-matrices (JADE) to separate the harmonic voltage and current fluctuation values at the common point (PCC) to obtain the mixing matrix., and then accurate the system harmonic impedance through the linear relationship. For data measurement and acquisition errors, data is cleaned using Pauta criteria. Considering that the harmonic impedance of the system will not be abrupt and the positive part is positive when the power system is running stably, after the system harmonic impedance is obtained, the data screening criterion of impedance value non-mutation and symbol discrimination is proposed. The simulation and field case calculation of the proposed method are carried out. The calculation results show that compared with the traditional method, the method can effectively improve the calculation accuracy.