To solve the detection problem in icing process, a thin film impedance sensor for atmospheric icing environment is proposed and evaluated. The sensor identifies rapid icing process and static ice layer as different conditions. A theoretical model of temperature-dependent impedance icing detection is built to describe the polarization effect. And the finite element simulation in the case of water film covering ice layer is carried out. The impedance curve distortion caused by phase mixing and temperature gradient during atmospheric icing is analyzed, and 9 related features are extracted. The evaluation experiments of the sensor involve 5 surface conditions, and totally 3143 impedance curves are obtained. The classification results of various models indicate that support vector machine achieves the highest recognition accuracy, reaching 93.1 %. And in atmospheric icing conditions, the RMSE of ice thickness measurement using neural network reaches 0.25 mm in the laboratory tests and 0.51 mm in the wind tunnel tests.