Abstract Background Within-sibship analyses show lower perinatal mortality after assisted reproductive technology (ART) compared with natural conception (NC), a finding that appears biologically unlikely. We investigated whether this may be attributed to bias from selective fertility and carryover effects. Methods Using data from national registries in Denmark (1994–2014), Finland (1990–2014) and Norway and Sweden (1988–2015), we studied 5 722 826 singleton pregnancies, including 119 900 ART-conceived and 37 590 exposure-discordant sibships. Perinatal mortality at the population level and within sibships was compared using multilevel logistic regression with random and fixed intercepts, respectively. We estimated selective fertility as the proportion of primiparous women with and without perinatal loss who had a second delivery, and carryover effects through bidirectional and crosswise associations. Results Population analysis showed higher perinatal mortality among ART conception compared with NC (odds ratio 1.21, 95% CI 1.13 to 1.30), whereas within-sibship analysis showed the opposite (OR 0.36, 95% CI 0.31 to 0.43). Primiparous women with perinatal loss were more likely to give birth again (selective fertility) and to use ART in this subsequent pregnancy (carryover effects), resulting in strong selection of double-discordant sibships with death of the naturally conceived and survival of the ART-conceived sibling. After controlling for conception method and outcome in the first pregnancy, ART was not consistently associated with perinatal mortality in the second pregnancy. Conclusions Whereas population estimates may be biased by residual confounding, within-sibship estimates were biased by selective fertility and carryover effects. It remains unclear whether ART conception contributes to perinatal mortality.