Background: Next generation sequencing (NGS) technology has been commonly applied to detect mtDNA mutations, which are reported to be strongly associated with cancers. However, several key challenges still exist in the bioinformatic analysis of mtDNA sequencing data, which greatly affect the detection accuracy of mtDNA mutations. Methods: Here, we systematically evaluated several key analysis procedures, including trimming, mapping and filtering, in mtDNA mutation detection of FFPE tissues, fresh tissues and plasma samples from cancer patients. Furthermore, the innovative bioinformatics pipeline integrating a newly-developed filtering algorithm was established.Results: We found that trimming procedure was essential for improving mtDNA mapping performance in plasma but not tissue samples. Mapping with rCRS-hg19 reference was strongly suggested for mtDNA mutation detection in plasma samples due to the extreme abundance of NUMTs. In addition, our results showed that the setting of 3 mismatches was most appropriate for mtDNA mutation calling. More importantly, we revealed the presence of a negative logarithmic relationship between mtDNA site sequencing depth and minimum detectable mutation frequency and thus build up an innovative and efficient filtering strategy to increase the accuracy and sensitivity of mutation detection. Finally, we verified that higher sequencing depth was required for PCR-based than capture-based enrichment strategy.Conclusions: Collectively, we established an innovative data analysis strategy, which is of great significance for improving the accuracy of mtDNA mutation detection for different types of tumor samples.