It is difficult to accurately calculate the harmonic impedance of the system due to background harmonic fluctuations. In this paper

three types of data filtering criteria to eliminate the influence of background harmonics are proposed based on the dominant fluctuation

linearity

and harmonic impedance non-mutation characteristics. The current data at the point of common coupling（PCC）is segmented

the data fluctuation extent is judged by the current variance in each segmentation of data and the data is

in accordance with the different fluctuation extent

is divided into three categories. The first type of the data judges the linearity of its voltage and current and the data meeting the linear threshold is considered to have the clustering feature. The harmonic impedance of the system of this type of data is calculated by the use of complex least square method. The calculation result is compared each other in chronological order

thereby judging its difference extent and taking out the difference. The proposed method and the data filtering method are compared and analyzed. It is shown by the simulation and actual case that the method in this paper is more accurate over the other methods.