For the reinforcement problem of three-dimensional rock slopes under the Hoek-Brown criterion, a non-discrete multi-objective optimization design method for anti-slide piles combining mechanical analysis and the response surface method was proposed. The stability model for three-dimensional pile-reinforced slopes was established using the upper bound limit analysis method. Artificial neural networks were adopted to establish the response surface model between the safety factor and the design and strength parameters to quickly obtain a large number of safety factor samples required for reliability evaluation. Furthermore, taking the reinforcement cost and slope reliability as optimization objectives, the Pareto front was directly solved by non-discrete point selection in the continuous design variable space. The case analysis results show that, compared with the discrete design method, the non-discrete design can more comprehensively explore the feasible design domain and obtain more optimized design schemes; under the minimum cost and knee point decision strategies, the costs of the non-discrete design decrease by approximately 18% and 14%, respectively; the pile spacing, pile diameter, and pile position all exhibit significant nonlinear effects on the slope reliability index, and the reinforcement effect is optimal when the piles are placed in the upper part of the slope face.