Abstract:An assumption about the failure width of failing soil mass usually should be presented at first when slope stability is evaluated by three-dimensional factor of safety or three-dimensional reliability index. But,as a result of this assumption,the three-dimensional factor of safety or reliability index isn¢t the minimal one,which can not represent the real safety level of slope stability. Actually there should be three-dimensional critical slip surfaces which give the highest failure probability if spatial variabilities of soil parameters are taken into consideration. The minimal three-dimensional reliability index and the corresponding critical slip surface are investigated and illustrated. This makes it possible to get three-dimensional reliability index of soil slope by means of optimization techniques. Genetic algorithm(GA) is good for finding the global optimum solution of problem searching for the critical three-dimensional surface of soil slope. So,the model of spatial variabilities of soil parameters is adopted and the GA is used to search for the reliability index directly,then the three-dimensional reliability analysis method for slope stability,based on the GA,is proposed in combination with Bishop¢s simplified method. The minimal three-dimensional reliability index,which can represent the real safety level of slope stability,and the corresponding three-dimensional slip surface are achieved without assumptions about the width of failing soil mass. This method is proven to be efficient by the examples.