In most species mitochondrial DNA (mtDNA) is inherited maternally in an apparently clonal fashion, although how this is achieved remains uncertain. Population genetic studies show not only that individuals can harbor more than one type of mtDNA (heteroplasmy) but that heteroplasmy is common and widespread across a diversity of taxa. Females harboring a mixture of mtDNAs may transmit varying proportions of each mtDNA type (haplotype) to their offspring. However, mtDNA variants are also observed to segregate rapidly between generations despite the high mtDNA copy number in the oocyte, which suggests a genetic bottleneck acts during mtDNA transmission. Understanding the size and timing of this bottleneck is important for interpreting population genetic relationships and for predicting the inheritance of mtDNA based disease, but despite its importance the underlying mechanisms remain unclear. Empirical studies, restricted to mice, have shown that the mtDNA bottleneck could act either at embryogenesis, oogenesis or both. To investigate whether the size and timing of the mitochondrial bottleneck is conserved between distant vertebrates, we measured the genetic variance in mtDNA heteroplasmy at three developmental stages (female, ova and fry) in chinook salmon and applied a new mathematical model to estimate the number of segregating units (Ne) of the mitochondrial bottleneck between each stage. Using these data we estimate values for mtDNA Ne of 88.3 for oogenesis, and 80.3 for embryogenesis. Our results confirm the presence of a mitochondrial bottleneck in fish, and show that segregation of mtDNA variation is effectively complete by the end of oogenesis. Considering the extensive differences in reproductive physiology between fish and mammals, our results suggest the mechanism underlying the mtDNA bottleneck is conserved in these distant vertebrates both in terms of it magnitude and timing. This finding may lead to improvements in our understanding of mitochondrial disorders and population interpretations using mtDNA data.