ObjectiveTo evaluate the mother–child separation rate in twin pregnancies delivered in maternity units offering an appropriate level of neonatal care.MethodsJUMODA is a French, prospective, population‐based cohort study of twin deliveries, including 7998 women who gave birth in maternity units with appropriate levels of neonatal care based on thresholds for weeks of gestational age (wGA) and birth weight according to French guidelines: level I (≥36 wGA), IIA (≥34 wGA), IIB (≥32 wGA and ≥ 1500 g), and IIII (<32 wGA or < 1500 g). The primary outcome was mother–child separation, defined as the transfer of at least one twin or the mother to another hospital.ResultsMother–child separation occurred in 2.1% of pregnancies. This rate was significantly higher in level I (4.8%, 95% confidence interval [CI] 1.5–12.5) and IIA (3.4%, 95% CI 2.4–4.7) compared with level IIB (1.6%, 95% CI 1.1–2.3) and level III maternity units (2.1%, 95% CI 1.9–2.8). In level IIA units, the rate of mother–child separation was higher for babies born between 34 and 36 wGA (8.3%) than for those born at 36 wGA or beyond (1.7%). In level IIb, the rate of mother–child separation for babies born between 32 and 34 wGA (7.5%) was higher than for those born between 34 and 36 wGA (2.1%) and at 36 wGA or beyond (0.9%).ConclusionMother–child separation rates were low but differed by level of care. By using specific thresholds for twins to define levels of care, rather than data from singleton births, one‐fifth of mother–child separations could have been avoided.