Abstract: Research on the multi-level lot-sizing (MLLS) problem in discrete manufacturing industry is not only theoretically but computationally hard. The solution to MLLS problem plays a very important role in gaining economic effectiveness during production process. To avoid the decrease in search efficiency caused by prematurity, the repulsion operator was integrated into GA (RGA) to solve uncapacitated MLLS problem with assembly structure. Simulation tests were conducted introducing both GA and RGA for 6 groups of MLLS problems of different sizes, and the computational results showed that RGA is obviously superior to GA and that RGA is an effective method to solve MLLS problem.