Various typing schemes have been developed to identify strain-specific differences and track transmission dynamics in affected communities, with recent introduction of whole genome sequencing providing the most accurate assessment. Mycobacterial interspersed repetitive unit (MIRU) typing is a family of variable number tandem repeat schemes that have been widely used to study the molecular epidemiology of M. tuberculosis. MIRU typing was used in most well-resourced settings to perform routine molecular epidemiology. Instances of MIRU homoplasy have been observed in comparison with sequence-based phylogenies, limiting its discriminatory value. A fundamental question is whether the observed homoplasy arises purely through stochastic processes, or whether there is evidence of natural selection. We compared repeat numbers at 24 MIRU loci with a whole genome sequence-based phylogeny of 245 isolates representing three modern M. tuberculosis lineages. This analysis demonstrated extensive homoplasy of repeat numbers, but did not detect any evidence of natural selection of repeat numbers, at least since the ancestral branching of the three modern lineages of M. tuberculosis. In addition, we observed good sensitivity but poor specificity and positive predictive values of MIRU-24 to detect clusters of recent transmission, as defined by whole-genome single nucleotide polymorphism analysis. These findings provide mechanistic insight, and support a transition away from VNTR-based typing toward sequence-based typing schemes for both research and public health purposes.