In the selective assembly of complex mechanical products, there is the replication phenomenon of individual components and parts. This paper proposes an improved multi-objective optimization NSGA2(non-dominated sorting genetic algorithm-2) based on an elite reserved strategy of the offspring combining population evenness and crowding degree. With assembly qualified rate and assembly precision as the quality evaluation index, a multi-objective optimization model for selective assembly is established. The deficiencies of the local search capability of the NSGA2 algorithm is overcome by introducing the nearest neighbor search operator. Taking the assembly of a certain type of aircraft piston engines as an example with the optimization result represented by the Pareto boundary set, the results show that the diversity and astringency of the non-dominated solution sets (non-dominated solution sets) are obtained after the algorithm is improved.