Women who have had a child with mitochondrial DNA (mtDNA) disease need to know the risk of recurrence, but this risk is difficult to estimate because mutant and wild-type (normal) mtDNA coexist in the same person (heteroplasmy). The possibility that a single sample may not reflect the whole organism both impedes prenatal diagnosis of most mtDNA diseases, and suggests radical alternative strategies such as nuclear transfer. We used naturally occurring mtDNA variants to investigate mtDNA segregation in placenta. Using large samples of control placenta, we demonstrated that the level of polymorphic heteroplasmic mtDNA variants is very similar in mother, cord blood and placenta. However, where placental samples were very small (< 10 mg) there was clear evidence of variation in the distribution of mtDNA polymorphic variants. We present the first evidence for variation in mutant load, that is, mosaicism for mtDNA polymorphic variants in placenta. This suggests that mtDNA mutants may segregate in placenta and that a single chorionic villous sample (CVS) may be unrepresentative of the whole placenta. Duplicates may be necessary where CVS are small. However, the close correlation of mutant load in maternal, fetal blood and placental mtDNA suggests that the average load in placenta does reflect the load of mutant mtDNA in the baby. Provided that segregation of neutral and pathogenic mtDNA mutants is similar in utero, our results are generally encouraging for developing prenatal diagnosis for mtDNA diseases. Identifying mtDNA segregation in human placenta suggests studies of relevance to placental evolution and to developmental biology.