In recent years, frequency domain reflection method (FDR) has been widely used in defect detection of power cables. However, multi-reflection of traveling waves and the reflection coefficient spectrum of FDR are non-periodic signals, which leads to false peaks in the positioning spectrum of FDR, and this problem can easily lead to serious misjudgment of the location of defects. In this paper, a new method for power cable defect location based on Pisarenko harmonic decomposition is proposed. The linear difference model is used to fit the probability density distribution of reflection coefficient spectrum and construct an auto-regressive moving average (ARMA) model to replace the fast Fourier transform (FFT) in the traditional FDR. The proposed method can effectively overcome the influence of attenuation, multi-reflection and dispersion effect on cable defect location. Then, Pisarenko harmonic decomposition method is used to estimate all frequencies in the reflection coefficient spectrum and further complete the accuracy location of defects. Finally, the proposed method is validated through simulation data and on-site data from a 58 km submarine cable. The results show that the proposed method can not only weaken the influence of false peaks on defect location, but also improve the location resolution.