Estimates statistics of the first two power spectrum moments from the pulse pair covariance are analyzed. The input signal is assumed to be colored Gaussian and the noise, white Gaussian. Perturbation formulas for the standard deviation of both mean frequency and spectrum width are applied to a Gaussian shaped power spectrum, and so is a perturbation formula for the bias in the width estimate.Mean frequency estimation from interlaced pulse pairs is presented. Throughout this study, estimators from independent, spaced, and contiguous pulse pairs are compared to provide a continuum of statistics from equispaced tightly correlated to statistically independent pulse pairs. Spectral moment estimation by covariance argument techniques is increasing as an important signal processing tool in the analysis of pulse Doppler echoes from distributed scatterers [1]. Hyde and Perry [2] suggested a version of the covariance technique for first moment estimation, and Woodman and Hagfors [3] used it to measure mean ionospheric motions. Independently, Rummler had proposed estimators of both mean and variance from pulse pair bursts [4], [5] . Hofstetter [6] and Miller and Rochwarger [7] established maximum likelihood properties of these estimators. Furthermore, Miller and Rochwarger analyzed the first and second order statistics of the estimators. The probability density of the mean frequency for a large number of independent pairs can be found in [3], while Berger [8] calculated the density for any arbitrary number of independent pairs. To process independent echo sample pairs requires much longer observation time; hence it has been suggested [4] that independence in a uniform pulse train can be created by changing the carrier frequency on altemate pairs. This, of course, considerably complicates the system's hardware. Moment estimation from contiguous pairs with constant carrier frequency is the most direct technique to implement on conventional radars. Arithmetic units that calculate spectral momnents using the fast Fourier transform algorithm also can be used. Although these units rapidly calculate the spectrum and its moments, all echo samples comprising the time series must be stored prior to calculations. When data fields consist of a thousand sample volumes along a radial and each volume generates a few hundred samples, the storage required normally exceeds a million bits. Estimates from contiguous pairs (with a common pulse among each two pairs) were analyzed by Benham et al. [9], and Berger and Groginsky [10] , who found the standard deviations of the mean and variance. In this paper the perturbation analysis of Benham et al. [Appendix A] is extended to include arbitrarily spaced pulse pairs. Under appropriate limiting conditions, it is shown how this more general situation reduces to previous results of Berger and Groginsky [10] (pair spacing equal to intrapair period), Miller and Rochwarger [7] (large spacing between pairs or uncorrelated pairs) and Lank et al. [11] (zero width spectrum or fully correlated signal sample...