Abstract: To assess the complex system-s risk probability while the number of the simples was limited, the small probability assessment modeling method by small simples was put forward, and the mapped relationship between the cumulating probability and the extremum of the critical parameter was approached by nonlinear function. The risk probability assessment mathematic model was developed, and the basic assessment model was obtained by using nonlinear regress. The offered simples can not completely accorder with the assessment model, so the error was regarded as the object function, and the optimum probability assessment model was presented. In order to optimize the model-s distribution parameter, the improved genetic algorithm was adopted. The applied example of flight safety assessment was showed, and the flight risk probability was calculated by the limited simples. The precise is improved by the optimum approach function.