Abstract: By using binomial smoothing and Morlet wavelet transform, we have researched the interannual/interdecadal variabilities on Nanpanjiang Basin in the major flooding season.Meanwhile, predictors about correlation surpassing 99% have been selected, on this basis, the forecasting model of optimal subset regression on Nanpanjiang Basin in the major flooding season has been established.The results show that firstly, from the linear trend, the precipitation of Nanpanjiang Basin in the major flooding season from 1961 to 2014 was decreasing, and the decreasing trend is 2.7mm yearly.Secondly, in interdecadal variability, the main period is quasi 14 and 18 years, but in interannual variability, the main period is quasi 3, 6 and 9 years, and interannual variability presents interdecadal variances.What is more, the consistent symbol rate of precipitation anomaly percentage observations and regression values of optimal subset regression is 70% ; regression accuracy is 87.5%; the consistent symbol rate of precipitation anomaly percentage observations and predicted values is 71.4%, and the forecast accuracy is 85.7%.It shows that the precipitation trend prediction level is sound.