A statistical verification of a number of forecast icing algorithms is presented. The algorithms are based on gridded forecasts of temperature and humidity from the global model of the European Centre for Medium-range Weather Forecasts. This study was undertaken with the use of a database of pilot reports issued from the Winter Icing and Storm Program and collected between 1993 and 1995 over the USA. The results are presented both in terms of percentage of detection and in terms of total surface of the forecast. They highlight the difficulty of having to make a choice between those two criteria. Taking into account vertical velocity in the algorithms reduces the percentage of detection, but the area where icing is forecast decreases to an even higher extent.