In order to study the influence of hydrological time-scale on the forecasting accuracy of medium- and long-range hydrological forecasting, the nearest neighbor bootstrapping regressive model and combination model based on wavelet analysis were used to forecast runoff series of different time scales in the Yangtze River Basin. The daily runoff series of Gaochang, Cuntan, Yichang, Luoshan, Hankou and Datong (the 6 key sections of the Yangtze River) from 1980 to 2012 were converted into ten time scales as three days, week, ten days, half month, monthly, bimonthly, quarterly, half a year, nine months and yearly, and used by two models to simulate and forecast. The results show that with the increase of the forecasting time scale, forecast accuracy tends to decrease first and then increase.