A method is presented for validation of space-based polarization measurements. The method is based on a determination of limits in the (fractional) polarization of reflected solar light as measured by the GOME instrument. These empirical limits were used to study the instrument performance during the first 5.5 years of GOME operation (1996--2001), which revealed a wavelength and viewing angle-dependent degradation of the GOME polarization measurement devices. Degradation correction factors are derived under the assumption that these polarization limits remain constant in time. Using results from earlier GOME studies on clouds, Lambertian equivalent reflectance, and geolocations, it is shown that these limits correspond to cloud-free scenes with minimal aerosol load and minimal surface albedo in the case of highly polarized observations, or to clouded scenes in the case of unpolarized observations. The polarization state of the atmosphere corresponding to observations of extremely highly polarized scenes is verified using vector radiative transfer calculations.