Dowry is often adduced as an explanation of son preference in India, but there is little evidence that dowry motivates son-preferring behaviours. On the premise that gold is an integral part of dowry, we use variation in gold prices to investigate this. First, we exploit a sharp unexpected rise in the price of gold in 1980 and, using a difference-in-discontinuities design, find that the gold price hike is mirrored in an increase in girl relative to boy mortality in the neonatal and infant period. We also find that surviving girls are shorter. Second, using monthly time series data for 35 years, we again find that cyclical variation in gold prices is reflected in excess girl mortality and, since the introduction of prenatal sex determination technology, in the sex ratio at birth. This constitutes the first evidence that dowry costs lead parents to eliminate foetal and newborn girls, and on a scale much larger than "dowry deaths" amongst married women which have been the subject of public attention.