Abstract: To study the relationship between rainfall and runoff in the upper reaches of Changjiang River Basin, this paper firstly constructs a set of rainfall and flood field division algorithms. Based on this algorithm, the mean area rainfall data and runoff data of main subbasins of the upper reaches of Changjiang River were divided into independent flood events. Then characteristic indices of rainfall and flood of each flood event were calculated and stored into the index database. On this basis, taking two most important flood indices, flood volume and flood peak flow, as dependent variables and rainfall index as independent variables, the partial least squares regression model was established to simulate the relationship between rainfall and runoff. The results show that the trained model shows high simulation accuracy in most watersheds, and the model can well simulate the relationship between rainfall and runoff. Prediction of flood index through this model can provide scientific reference for the flood forecast in the upper reaches of Changjiang River Basin.