In order to better carry out the road icing prediction and early warning services, the hourly observation data of traffic meteorological stations in the high incidence area of road icing in Gansu Province (the east of Wuwei, Gansu) were used to analyze the spatial and temporal distribution characteristics of road icing, explore the correlation between road icing and meteorological factors, and construct the road icing warning model by using Logistic regression method and neural network algorithm. The results showed that road icing in Gansu Province occurred mainly in winter (from December to February of the following year), and the frequency of road icing was higher from 00:00 to 10:00 and from 22:00 to 23:00. Logistic regression model and neural network model had high prediction accuracy for non-icing events, with 91.9% and 96.2%, respectively. For the occurrence of icing events, the prediction accuracy of Logistic regression model was low, at 31.6%, while that of neural network model could reach 44.6%, indicating that the two models had certain indicative significance for road icing warning, and the prediction effect of neural network model was better than that of Logistic regression model.