Abstract: To solve the problem of single-objective multi-modal expected value programming with unknown noisy distribution, a multi-objective immune optimization algorithm based on immune response principles was proposed. By means of a random function used for checking whether a candidate solution was a locally optimal solution, the expected value problem was converted into a multi-objective expected value programming problem. Moreover, some relations between the original problem and the transformed problem were studied, and a conclusion that an efficient solution must be an optimal solution under certain conditions was developed. Relying upon the idea of sample average approximation, the multi-objective programming was further transformed into an approximate model with variable sampling sizes. Based on the metaphors of clonal selection and immune memory, one such optimization approach was obtained to deal with the approximation model. It searched high-quality individuals toward some regions which the optimal solutions existed, depending on several main modules: recursive non-dominated sorting, adaptive sampling, adaptive proliferation, and adaptive mutation. By comparison with the multi-objective optimization algorithms, the simulation results show that the proposed algorithm with strong noise suppression can achieve averagely about a seventy percent increase in the number of optimal solutions found for the high-dimensional benchmark problem and a twenty percent increase for the low-dimensional benchmark problem; it can gain the stable search effect and has the prominent advantage in solving multiple optimal solutions.