Heteroplasmic mutations in mitochondrial DNA (mtDNA) play critical roles in mitochondrial disease, aging, and cancer. Recently, next-generation sequencing (NGS) has been widely used to detect mtDNA mutations for diagnosis and monitoring of the above-mentioned diseases. However, little attention is paid on inherent cross-contamination generated during mtDNA capture and sequencing of mixed samples, which may seriously reduce the detection accuracy of mtDNA heteroplasmic mutations. In this study, a novel sequencing strategy based on a unique double-barcode design was established. The results showed that when single barcode-based analysis strategy was used, cross-contamination level of 20 DNA samples ranged from 0.27% to 11.90% on HiSeq 2500 and from 0.93% to 17.70% on HiSeq X ten, whereas double barcode-based strategy could effectively eliminate cross-contamination. Moreover, the data indicated that cross-contamination was mainly derived from capture process and was significantly affected by different NGS platforms. In addition, contamination level was negatively related to sequencing depth. Moreover, cross-contamination significantly increased the false-positive calling of mtDNA heteroplasmic mutations and remarkably affected the heteroplasmy level of mtDNA mutations. In contrast, cross-contamination had no notable effect on classification of mtDNA haplogroup. Taken together, our novel double barcode-based sequencing strategy is effective in eliminating cross-contamination, enhancing the detection accuracy of mtDNA NGS, and improving its application in diagnosis or monitoring of diseases associated with mtDNA mutations.