Abstract: By using the hetero-associative memory of nerve-network and the similarity of the error that can enlarge the real-time information,a comprehensive real-time correction method is presented.An auto-regressive(AR)model and a comprehensive correction method are both made for real-time correction of 11 floods in Qilijie Basin.The results show that a comprehensive correction method is more accurate and the forecast lead period is not lost,comparing with the AR model.The results of using the comprehensive correction method for testing the nearly 50 floods in six tributary basins of Minjiang Basin are satisfactory.