<p>We present a detection method and measurements of low-friction formations on road surfaces using a monostatic polarimetric radar installed on a vehicle in motion. Polarimetric parameters (PP) such as entropy, depolarization, and the proportion of surface-scattering are used to identify the surface and classify it as dry, wet, or icy. The polarimetric parameters are calculated from the eigenvalues (EV) of the measured covariance/coherence matrix. Due to the low contrast in the polarimetric parameters between icy and dry surfaces, the classification of the surface becomes challenging in the presence of irregularities on the surface that can be interpreted as ice. To minimize the false alarms, in addition to the PP, we use the sum of the non-normalized eigenvalues, which is a measure of the total scattered power in all polarization components. The sum of the EV has better contrast compared to the PP and helps in reducing the negative effect of natural surface irregularities. To classify the surface, we calculate the adaptive mean value and variance of the PP and non-normalized EV and compare each measurement to a threshold. The surface is classified after a certain combination of parameters exceeds their threshold value. We demonstrate how each of the parameters is affected by an icy/wet surface compared to the same surface in dry conditions. In our measurements, we are able to detect ice patches with a false alarm below 2% with a simple classification criterion combining some of the PP with the sum of the EV.</p>