The paper suggests to treat the infrared reflectivity spectra of single crystal perovskite relaxors as fine-grained ferroelectric ceramics: locally frozen polarization makes the dielectric function strongly anisotropic in the phonon frequency range and the random orientation of the polarization at nanoscopic scale requires to take into account the inhomogeneous depolarization field. Employing a simple effective medium approximation (Bruggeman symmetrical formula) to dielectric function describing the polar optic modes as damped harmonic oscillators turns out to be sufficient for reproducing all principal features of room temperature reflectivity of PMN. One of the reflectivity bands is identified as a geometrical resonance entirely related to the nanoscale polarization inhomogeneity. The approach provides a general guide for systematic determination of the polar mode frequencies split by the inhomogeneous polarization at nanometer scale. In recent years, there has been enormous effort in studying single crystals with intrinsic nanoscopic inhomogeneity, since such materials often show a very interesting properties. It was even proposed that the clustered, inhomogeneous states encountered for example in high-Tc cuprates, CMR manganites, nickelates, cobaltites, diluted magnetic semiconductors or ferrolectric relaxors, should be considered as a new paradigm in condensed matter physics. 1 In case of relaxors, the peculiar dielectric properties of relaxor materials were associated with the presence of small polar clusters -so called polar nano-regions (PNR's)-already in the pioneer work of Burns and Dacol 2 . However, because of their small size and random nature, we still lack a clear understanding of their size distribution, thickness and roughness of their boundaries, their connectivity, shape anisotropy, distribution of the associated dipolar moments, their fractal self-similarity, their dynamics, and so on. PNR's are often represented as small islands submerged in a nonpolar matrix, possibly appearing and disappearing again in time. On the other hand, the recent piezoelectric scanning microscopy investigations 3 of the surface of PbTiO 3 -doped relaxors rather invoke a picture of a fine, hierarchical and essentially static "nanodomain" structure. It strongly suggests that the common perovskite relaxors are actually quite densely filled by quasi-static polar nano-regions, and that the former picture with prevailing non-polar matrix can perhaps be appropriate only at high temperatures around the so called Burns temperature. 2,4