In most existing cases of interpolated fast Fourier (IpFFT), only the first and second highest spectral lines around the spectral peak are used. To decrease estimation variance, the composite IpFFT on four consecutive spectral lines around the peak is studied. Four consecutive lines render three estimators. The optimal coefficients at the minimum variance are deduced after the weighted average of the estimators. Theoretical analysis shows that the variance of the proposed composite correction is lower than that of the Quinn, and the greatest improvement is achieved under the coherent samplingcondition, with the variance deduction down to 4/9 of the Quinn′s. This theoretical analysis is validated through numerical simulation. It is shown that the empirical variance fits the theoretical expression when the signal-noise ratio (SNR) is high (50 dB) and the sampling is non-coherent. When the SNR is down to 0 dB, the empirical variance deviates from the theoretical expression no more than 25%.