Abstract: Precipitation of short-term climate prediction is a very complex and important research topic. In order to improve the predictive capability, particle-swarm-optimization based neural network reasoning models were established for Xuancheng, Anhui with summer precipitation prediction. The monthly data of 74 circumfluent eigen values, the monthly data of sea surface temperature, the monthly data of 500 hPa height from 1959 to 2011 to choose forecast factor, and principal component analysis method to extract the main information in the sample data for the comprehensive factor. 2007-2011 Xuancheng summer precipitation forecast verification results showed, particle swarm optimization artificial neural network convergence speed was fast, decrease in the number of iterations, and predicted the average absolute error was 66.5 mm, the absolute value of the average relative error of 10.5%, accurate prediction and had a good application prospect.