The present work proposes an optical polarization based reflection technique for determining surface roughness of metal surfaces. A linearly polarized light beam is used to illuminate the sample surface and two intensity data frames are recorded with specific orientations of an analyzer. An algorithm is developed to combine the pair of intensity data frames captured with this simple setup. Assuming that the refractive index of the sample is known and constant throughout the sample, the phase difference between the transverse electric and transverse magnetic components of the light reflected from the surface preserve the information about surface roughness in terms of local angular tilts, which is subsequently converted to axial deviations about the nominal surface. © 2021