Based on the comprehensive analysis of the correlation between visibility and various meteorological elements, the short-term prediction model for highway visibility was studied by using the visibility, temperature, humidity and wind strength data from Anhui Provincial Highway Visibility Observation Station Network. Experiments were conducted with humidity, temperature, average wind speed, instantaneous wind speed, and maximum wind speed as the input layer of the BP neural network, and the visibility as the output layer. The results show that the overall deviation of the experimental data is within the acceptable range. For sequential test samples, the tests whose relative error within 20% account for 68.6% of the total number of tests. In each random sample test, the BP network simulation output shows a good correlation with the test sample, with a correlation coefficient between 0.6 and 0.8. The low-visibility random sample test results show thatthe root mean square error between the model output value and the sample value is in the range of 700-850 m, with little amplitude of variation, indicating that the neural network algorithm has relatively high stablility.