Abstract: Fast and effectively estimating the frequency of sinusoidal signals plays an important role in fields such as radar signal processing, communication, power grid harmonic detection and electronic countermeasures. The existing all-phase interpolation iterative frequency estimation usually uses the spectral line interpolation of half frequency resolution on both sides of the all-phase FFT (apFFT) spectral center line directly. When the frequency of the interpolated spectral line is much different from the true frequency, the amplitude of the spectral line is small, resulting in a large estimation error at low signal-to-noise ratio. In this paper, an improved all-phase interpolation method is proposed, which uses the apFFT spectrum for bispectral line interpolation, and uses the two spectral lines with the highest amplitude as the amplitude ratio to derive the relationship between frequency offset and amplitude ratio, thereby obtains the frequency estimation. Simulation results show that compared with the all-phase interpolation iterative algorithm, the proposed method has better stationarity and smaller error in estimation root mean square error under different frequency biases, and the frequency estimation performance approaches the Cramer-Rao lower bound at high signal-to-noise ratio.