The invention provides a nonlinear runoff probability prediction method. The method is characterized by comprising the following steps that 1, unit lines which are generated by net rainfall within the drainage basin time frame (j, j+delta t) and submit to distribution G(.); k unit line impact factors tau j,1, tau j,2,..., tau j,k within the time frame (j, j+delta t) are selected, and a correlation H(.) between the unit line impact factors and theta j is built; 2, runoff of the net rainfall generated at the moment T within the time frame (j, j+delta t); 3, it is assumed that true runoff of the net rainfall generated at the moment T within the time frame (j, j+delta t) shows normal distribution and fluctuates vertically relative to a predicted value; it is further assumed that the true runoff, the predicted value and the impact factors tau j,1, tau j,2,..., tau j,k conform to the formula shown in the description; 4, the true runoff at the moment T is obtained; and distribution which QT meets is obtained; 5, a likelihood function of the flood is obtained according to the drainage basin, a likelihood function of multiple floods is calculated, and the maximum likelihood method is used for estimating parameters; 6, according to the obtained parameters and the runoff models built in the steps 1-4, prediction is performed.