Calculation of forest fire probability is a complex issue concerning weather, tree species, geography conditions and human activities. The impact of weather parameters on fire has been one of the hot spots in the forest fire study. In this paper, the effect of 5 daily weather parameters on forest fire probability is investigtated. The 5 parameters are average humidity, precipitation, average wind speed, average temperature and sunshine time. Firstly, the correlation between each single parameter and fire probability is analyzed with a 24ayer BP neural network. The little value of MSE indicates ANN method gets close results with actual correlation. Secondly, the multi-correlaiton between fire probability and the 5 weather parameters is studied with a 3-layer BP network. From the 466 samples, 26 ones are randomly selected as test set, others as training set. The training MSE of BP network becomes smaller than 10^-6 after 5 000 epochs. For the test set, the relative error is less than 9.9%. It is indicated from the results that there are steady correlaiton between fire probability and weather parameters, and the BP network is a practical method in fire risk analysis. The study has practical implicaitons for forest fire risk prediction and the results can act as a basic data in forest fire proteciton.