The ‘insurance ova’ hypothesis (Anderson, 1990) views dizygotic twinning as a by-product of selection for multiple ovulation which sometimes – in error – results in the birth of twins. From this viewpoint, polyovulation is a mechanism which reduces the risks of fertilization failure or embryo defect/mortality. If DZ twin births are a ‘side-effect’ of a mechanism which compensates for defective embryos one might predict that embryo defect rates and twinning rates will covary. This prediction is tested using national-level data on twinning rates and rates of trisomy-21 (Down's syndrome), and a strong positive correlation is found, even when controlling for maternal age. One suggestion that follows from this finding is that intra- and interpopulation variation in both twinning rates and Down's syndrome rates may result, in part, from individual variation in pre-implantation rejection of embryos in the very early stages of pregnancy. In this paper it is proposed that the ‘insurance ova’ explanation for twinning in humans could be expanded to incorporate a model of rejection of anomalous embryos, be they anomalies of number or type. Variation in the efficiency of an embryo rejection mechanism, combined with variation in frequency of polyovulation, would have consequences for individual reproductive success.