Abstract We propose a data adaptive spectral estimation algorithm which is suitable for nonstationary estimation situations. This algorithm is based on the conventional Fourier transform of the estimated autocorrelation function. The data adaptive feature is implemented into the autocorrelation function estimation. The algorithm is computationally efficient due to its recursive nature. Its frequency tracking performance is tested against another adaptive algorithm based on the frequently used least mean square algorithm (LMS) of Widrow and Hoff (1960). The two algorithms demonstrate similar performance in many situations. Computer simulations indicate that, when applied to a signal composed of two sinusoids with different power levels, the proposed algorithm tracks the lower-powered sinusoid better than the LMS algorithm.