The cepstrum thresholding approach for variance reduction is applied to estimate the spectrum of mesospherestratospheretroposphere radar data, which are complex. Unlike in a real data case, where the cepstral values are real, for complex signals, cepstral values will be complex. We propose two different thresholding approaches: 1) Applying a threshold independently for Real and imaginary values of cepstral estimates and 2) threshold application to the modulus of cepstral estimates. Reductions of the total variance of the estimated spectra using the aforementioned two threshold approaches have been tested with simulated data. Finally, the algorithm has been applied to the practical atmospheric radar data to estimate the wind parameters through the computation of Doppler shift. Results have been validated using simultaneous Global Positioning System sonde data. © 2006 IEEE.