Mitochondria frequently carry different DNA—a state called heteroplasmy. Heteroplasmic mutations can cause mitochondrial diseases and are involved in cancer and aging, but they are also common in healthy people. Here, we study heteroplasmy in 96 multigenerational healthy families. We show that mothers effectively transmit very few mitochondrial DNA to their offspring. Because of this bottleneck, which intensifies with increasing maternal age at childbirth, mutation frequencies can change dramatically between a mother and her child. Thus, a child might inherit a disease-causing mutation at high frequency from an asymptomatic carrier mother and might develop a disease. We also demonstrate that natural selection acts against disease-causing mutations during germline development. Our study has important implications for genetic counseling of mitochondrial diseases.Heteroplasmy—the presence of multiple mitochondrial DNA (mtDNA) haplotypes in an individual—can lead to numerous mitochondrial diseases. The presentation of such diseases depends on the frequency of the heteroplasmic variant in tissues, which, in turn, depends on the dynamics of mtDNA transmissions during germline and somatic development. Thus, understanding and predicting these dynamics between generations and within individuals is medically relevant. Here, we study patterns of heteroplasmy in 2 tissues from each of 345 humans in 96 multigenerational families, each with, at least, 2 siblings (a total of 249 mother–child transmissions). This experimental design has allowed us to estimate the timing of mtDNA mutations, drift, and selection with unprecedented precision. Our results are remarkably concordant between 2 complementary population-genetic approaches. We find evidence for a severe germline bottleneck (7–10 mtDNA segregating units) that occurs independently in different oocyte lineages from the same mother, while somatic bottlenecks are less severe. We demonstrate that divergence between mother and offspring increases with the mother extquoterights age at childbirth, likely due to continued drift of heteroplasmy frequencies in oocytes under meiotic arrest. We show that this period is also accompanied by mutation accumulation leading to more de novo mutations in children born to older mothers. We show that heteroplasmic variants at intermediate frequencies can segregate for many generations in the human population, despite the strong germline bottleneck. We show that selection acts during germline development to keep the frequency of putatively deleterious variants from rising. Our findings have important applications for clinical genetics and genetic counseling.