Although several optimal mycobacterial interspersed repetitive units-variable number tandem repeat (MIRU-VNTR) loci have been suggested for genotyping homogenous Mycobacterium tuberculosis, including the Beijing genotype, a more efficient and convenient selection strategy for identifying optimal VNTR loci is needed. Here 281 M. tuberculosis isolates were analyzed. Beijing genotype and non-Beijing genotypes were identified, as well as Beijing sublineages, according to single nucleotide polymorphisms. A total of 22 MIRU-VNTR loci were used for genotyping. To efficiently select optimal MIRU-VNTR loci, we established accumulations of percentage differences (APDs) between the strains among the different genotypes. In addition, we constructed a minimum spanning tree for clustering analysis of the VNTR profiles. Our findings showed that eight MIRU-VNTR loci displayed disparities in h values of ≥0.2 between the Beijing genotype and non-Beijing genotype isolates. To efficiently discriminate Beijing and non-Beijing genotypes, an optimal VNTR set was established by adding loci with APDs ranging from 87.2% to 58.8%, resulting in the construction of a nine-locus set. We also found that QUB11a is a powerful locus for separating ST10s (including ST10, STF and STCH1) and ST22s (including ST22 and ST8) strains, whereas a combination of QUB11a, QUB4156, QUB18, Mtub21 and QUB26 could efficiently discriminate Beijing sublineages. Our findings suggested that two nine-locus sets were not only efficient for distinguishing the Beijing genotype from non-Beijing genotype strains, but were also suitable for sublineage genotyping with different discriminatory powers. These results indicate that APD represents a quantitative and efficient approach for selecting MIRU-VNTR loci to discriminate between divergent M. tuberculosis sublineages.