In order to further improve the estimation precision of sinusoid frequency in additive white Gaussian noise background, a new frequency estimation algorithm of sinusoidal signals based on interpolated Fast Fourier Transform (FFT) was proposed. Firstly, zeros of length N were padded to the sinusoid sampled data of length N in the time domain. Next, 2 N-point FFT was performed and the coarse estimation was made by searching the location of the discrete spectrum line with maximum amplitude. Finally, the fine estimation was made by utilizing the spectrum line with maximum amplitude and two sample values of Discrete-Time Fourier Transform (DTFT) of the original signal on the left and right side of the maximum spectrum line. Simulation results show that the root mean square error of the proposed estimator is close to the Cramer-Rao lower bound when the signal frequency locates anywhere between two neighboring FFT discrete spectral lines and the performance is stable. The estimation precision is higher than Candan estimator, Fang estimator, Rational Combination of Three Spectrum Lines (RCTSL) estimator and Aboutanios estimator. The proposed estimator also has lower signal-to-noise ratio threshold than the existing estimators.