In this paper we introduce a new radar-imaging technique, called polarization coherence tomography (PCT), which employs variation of the interferometric coherence with polarization to reconstruct a vertical profile function in penetrable volume scattering. We first show how this profile function can be efficiently represented as a Fourier-Legendre series, with tomographic reconstruction reducing to estimation of the unknown coefficients of this series from coherence data. We then show that we can linearize this inversion by using a priori knowledge of two parameters, namely, volume depth and topographic phase. We further propose a new algorithm based on polarimetric interferometry to estimate these two from the data itself. To assess stability, we investigate both the single- and dual-baseline conditioning of the associated matrix inversion and then concentrate on the single-baseline case to demonstrate that for sufficient multilooking (around 50), stable retrievals of profiles can be obtained in the presence of coherence noise. Finally, we apply the technique to simulated L band coherent radar data to demonstrate its potential for new applications in radar remote sensing.