A partial sequence of the mitochondrial cytochrome oxidase subunit I (COI) gene is widely used as a molecular marker for species identification in animals, also termed a DNA barcode. However, the presence of more than one sequence type in a single individual, also known as heteroplasmy, is one of the shortcomings of barcode identification. In this study, we examined the extent and divergence of COI heteroplasmy, including nuclear-encoded mitochondrial pseudogenes (NUMTs), at the genomic-DNA level from 13 insect species, including four individuals of orthopteran Anapodisma miramae. Furthermore, a long fragment of mitochondrial DNA (~13.5 kb) and cDNA from A. miramae were used as a template for COI PCR to compare the patterns of heteroplasmy between DNA sources and to investigate a possible way to avoid ambiguity in DNA barcoding. When multiple numbers of clones originated from genomic DNA were sequenced, heteroplasmy was prevalent in all species (3~16 heteroplasmic copies), with a varying degree of maximum sequence divergence (<1% in 7 species, <4% in 3 species, <6% in 2 species and 2.15-8.03% in four A. miramae individuals). In five species, NUMTs also were observed when genomic DNA was used as a template. Long fragment DNA also is a source of heteroplasmic amplification, but the divergent haplotypes and NUMTs obtained in the genomic DNA-based PCR were not detected in A. miramae. On the other hand, cDNA was heteroplasmy-free, without NUMTs when multiple numbers of clones were sequenced. Consistently, one dominant haplotype was always obtained from the genomic DNA-origin clones in all species and also from the long fragment- and cDNA-origin clones of A. miramae. Furthermore, the dominant haplotype was identical in sequence, regardless of the DNA source. Thus, one possible solution to avoid the barcoding problem in relationship to heteroplasmy could be the acquisition of multiple numbers of barcoding sequences to determine a dominant haplotype that can be assigned as barcoding sequence for a given species.