Abstract: With application of neural network theory, a new method has been proposed for direct estimation of the safety factor for circular and wedge failure of slopes. Furthermore, a new algorithm called generalized learning algorithm is designed to learn better knowledge of estimation of the safety factors of slopes from cases. The learned knowledge is then used for extrapolating prediction of the safety factor of new slope, Compared with the safety factors predicted by the limit equilibrium and maximum likelihood method, the neural network method has high accouracy of extrapolating prediction and high ability of self-learning and processing uncertainty.