Abstract: In this paper, the power spectrum of theoretical correlation function and signal sample of the deterministic narrow band signal and the power spectrum of signal sample of the random narrow band signal are estimated by autoregressive (AR) and FET methods. Spectrum estimation of sodar signal is explained well by above results. When the distance between the two peaks in the power spectrum of sodar signal is small, the resolution of AR is inferior to FFT. When the distance between the two peaks in the power spectrum is large or th ere is single peak in power spectrum, the resolution and accuracy of AR is better than FFT. In all of these cases, the spectrum curve estimated by AR method is smooth and has low side lobe. Combinati on of both of the AR and FFT is a good approach to estimate the power spectrum of sodar signal.