Generalizing radar interferometry to three-dimensional space, we present a model which predicts the variation with respect to frequency of the phase of the cross-spectrum obtained between two separate antennas. The slope of this variation is proportional to the cosine of the angle between the horizontal wind and the interferometer baseline, and inversely proportional to the horizontal wind amplitude. The zero-frequency phase depends on the horizontal wind speed and direction, on the vertical wind velocity, and on the anisotropy of the backscattered power with respect to the zenith. The model also allows us to predict qualitatively the effects of specularity, beam shape, and random turbulence on the phase versus frequency variation. Assuming a spaced receiver setup with three baselines rotated by 120°, we present an implementation scheme of the radar interferometry technique which yields a three-dimensional wind estimate and a parameter characterizing the anisotropy of the echo power distribution. Finally, we argue that the radar interferometry horizontal wind measurements will suffer from limitations similar to those of the so-called apparent velocity as both techniques are related by Fourier transform, and that the vertical wind component will be unbiaised by anisotropy in the backscattered power.