Based on the measured data of annual runoff from 1972 to 2011, the annual runoff variation characteristics of 4 catchments in Shule River were analyzed by inclination trend analysis and Mann-Kendall catastrophe trend test, and the BP neural network and particle swarm optimization-neural network were used to simulate the annual runoff prediction. The results indicated that the annual runoff of Changmabao, Dangchengwan, Shuangtapu, and Panjiazhuang are increasing, the slope of annual runoff cumulative anomaly percentage is respectively 13. 87%, 4. 46%, 11. 57%, 10. 49% per decade, the annual runoff frequency occur break in the 2004, 1983, 2008 and 2010, respectively. The 25-year scale period is the main control period of the annual runoff in the Shule River Basin. The particle swarm-neural network is more accurate than the Bp neural network during annual runoff simulation. Meanwhile, the annual precipitation will increase from 15% to 25% on the basis of 40 years mean value, and the runoff will not change significantly.