Large disparities in upper mesospheric winds measured with different radar techniques have been attributed earlier to the possibility that MF radars are sensitive to the phase speed of waves, rather than to true winds. An alternative explanation attributes these disparities to biases due to horizontal gradients in the wind field. These biases are common to the Doppler beam swinging (DBS), spaced antenna (SA), and imaging Doppler interferometry (IDI) methods and arise because of spatial filtering of the wind field by the beam configuration. We develop a theoretical model for intrinsic frequency and vertical wave number spectra of DBS wind component estimates that includes the effect of spatial filtering on an ensemble of gravity waves. The model shows a strong enhancement of frequency components above a knee, at ~1 hour intrinsic period. A similar model for the SA experiment is inherently intractable. We suggest, instead, that the DBS model should also hold for the spectra of SA wind component estimates. The knee is clearly evident in recently observed SA wind component spectra, but is less pronounced and occurs at a frequency lower than the DBS model predicts. Better agreement is expected with a model refined to include the effect of Doppler shifts due to background winds. Occurrence of the knee in observed SA wind spectra implies that MF radars are sensitive to winds rather than to phase speed of atmospheric waves. ¿ American Geophysical Union 1996