Abstract：The data sampling frequency of the signals in digital substation is normally a fixed 10kHz. The sampled points in every period are 200, not an integer power of 2. Meanwhile, it is very difficult to perform synchronized sampling for the fundamental frequency is fluctuating in the prescribed scope. The Fast Fourier transform may not suitable for its harmonic analysis. The accuracy can’t meet the requirements in power system. The Arithmetic Fourier transform (AFT) is more appropriate for analyzing discrete signals due to the advantages such as simple computation, good parallelism and no limitation on sampled numbers. It requires non-uniform sampling points, so the uniform sampled signals should be interpolated when using AFT. The interpolation method AFT keeps using is the zero interpolation. It produces a negative effect on the accuracy of the harmonic analysis, which can’t satisfy the requirements in power system. This paper proposes a new interpolation method of AFT after comparing the accuracy performance of four interpolation methods, i.e. the zero interpolation、the first-order linear interpolation、the piecewise cubic hermit interpolation and the cubic spline interpolation. Finally, the cubic spline interpolation is selected to improve the accuracy due to its higher precision and better stability. The Matlab simulation results show that the new interpolation can meet the requirements of power system harmonic analysis, and provides new ways for harmonic analysis.