In this paper, we propose several improvements to the standard double anemometry method for ice detection. In the double anemometry method, the wind speed measurement from a heated reference anemometer is compared with that of an unheated anemometer. A lower measurement from the unheated sensor suggests the presence of ice. First, we propose using a wind speed difference (not a ratio), because anemometers should not deviate significantly at any wind speed under ice‐free conditions. Second, the threshold should vary with ambient temperature to account for cup anemometer slowdown caused by thickening bearing grease. Finally, sensitive thresholds should be used to overdetect ice and false events removed during postprocessing. We created an algorithm to automatically determine the required thresholds and tested it on seven wind turbines during a full winter at a cold climate site. When compared with ice thickness measurements from cameras, the algorithm was equal to or outperformed the manual double anemometry method across all seven turbines.