Frost forms on bridges in Iowa about thirty times per year and presents a potentially hazardous condition for motorists. Accurate frost forecasts allow roadway maintenance personnel to make timely applications of preventative or suppressant material and minimize environmental impact from fugitive chemicals. However, accurate predictions present a challenge to forecasters due to high spatial variability of key meteorological factors leading to frost. A series of models were developed through the use of an artificial neural network to forecast the parameters (air temperature and dew point, bridge surface temperature, and winds) needed to drive an algorithm for frost deposition on bridges. The neural network was trained on model output and observations for four observation sites from three cold seasons (1995-1998). The frost model was then tested on data from the 2001-2002 and 2002-2003 cold seasons in Ames, IA. The frost forecast was developed to be issued at 18 UTC (12 PM) daily for twenty minute intervals beginning at 00 UTC (6 PM) and ending at 15 UTC (9 AM) local time. Results show that the artificial neural network forecast method produces more accurate forecasts in the short term than model output statistics derived from the 6 AM local time run of one of the National Weather Service models. Over the forecast period, the artificial neural network displayed a bias for under forecasting dew point, air temperature and surface temperature. Despite these biases, it is shown that the use of an artificial neural network as a tool for forecasting meteorological parameters is possible given the appropriate data.