Abstract: How to determine the impact weights of the previous water and sediment conditions for the delayed response of bank-full discharge in alluvial river is a difficult question. Considering the complex variation in impact weights of previous water and sediment conditions, an improved convolution model for the previous water and sediment conditions and their impact weights is established for bank-full discharge, and a simple correlation analysis method for determining and optimizing the impact weights of previous water and sediment conditions is proposed. The former bank-full discharge delayed response model using moving average and rate equation methods are both unique impact weight cases of convolution model. Taking the Sanhuhekou station located in the Inner Mongolia reach of Yellow River as a case study, the variation of bank-full discharge is simulated using different modes. The results indicate that the order of model effect from high to low is convolution model, moving average method, and rate equation method, the deterministic coefficients are 0.906, 0.903, and 0.879 respectively; the Nash-Sutcliffe efficiency coefficients are 0.906, 0.899, and 0.874 respectively, caused by different impact weights of previous water and sediment conditions. So the model performance can be improved by selecting the impact weights of previous water and sediment conditions reliably. Analysis of the optimum impact weights of previous water and sediment conditions shows that, the bank-full discharge is influenced by the water and sediment conditions of current year and preceding 6—8 years, with a dual effect.