Massive parallel sequencing (MPS) can accurately quantify mitochondrial DNA (mtDNA) single nucleotide variants (SNVs), but no MPS methods are currently validated to simultaneously and accurately establish the breakpoints and frequency of large deletions at low heteroplasmic loads. Here we present the thorough validation of an MPS protocol to quantify the load of very low frequency, large mtDNA deletions in bulk DNA and single cells, along with SNV calling by standard methods. We used a set of well-characterized DNA samples, DNA mixes and single cells to thoroughly control the study. We developed a custom script for the detection of mtDNA rearrangements that proved to be more accurate in detecting and quantifying deletions than pre-existing tools. We also show that PCR conditions and primersets must be carefully chosen to avoid biases in the retrieved variants and an increase in background noise, and established a lower detection limit of 0.5% heteroplasmic load for large deletions, and 1.5 and 2% for SNVs, for bulk DNA and single cells, respectively. Finally, the analysis of different single cells provided novel insights into mtDNA cellular mosaicism.