We present an empirical approach to model the wing polarization of strong resonance lines. This procedure based on `last scattering approximation' (LSA) was developed by citet{ste80,ste82}, for coherent scattering in the laboratory frame. We generalize his empirical approach to handle partial frequency redistribution (PRD). We illustrate this approach by applying it to the Ca I 4227 Å line. The LSA approach is successful in reproducing the observed Stokes Q/I polarization, including the location of the wing polarization maxima and the minima around the Doppler core, but fails to reproduce the observed spatial variations of the far wing polarization in terms of magnetic field and PRD effects. This null result points in the direction of a non-magnetic origin, which may include local deviations from a plane-parallel stratification with an inhomogeneous solar atmosphere.