Selective assembly method has progressed significantly for the past few years to become a valuable tool for improving quality in product assembly process where the required assembly precision is very high. In traditional selective assembly process, if the mating parts with non-normal distribution are grouped and assembled, many assembled products fail to meet the assembly precision requirement and thereby being identified as unacceptable to be scrapped. This paper proposes an approach by applying improved grouping selective assembly scheme to a ball-bearing assembly, to reduce the surplus parts and hence to improve acceptance rate of assembled products. A solving algorithm is presented based on genetic algorithm (GA), where the elitist strategy is integrated to improve the convergence of the algorithm, and the simulation is utilized to give better insight into the optimization process. Finally, some numerical experiments in different cases are conducted, which demonstrate that the proposed approach outperforms traditional selective assembly method in generating solutions with maximum assembly acceptance rate.