Objectives To identify obstetrical subgroups in which (1) the caesarean delivery (CD) rate may be reduced without compromising safety and (2) CD may be associated with better perinatal outcomes. Design A multicentre cross-sectional study. Setting 19 hospitals in the USA that participated in the Consortium on Safe Labor. Participants 228 562 pregnant women in 2002-2008. Main outcome measures Maternal and neonatal safety was measured using the individual Weighted Adverse Outcome Score. Methods Women were divided into 10 subgroups according to a modified Robson classification system. Generalised estimated equation model was used to examine the relationships between mode of delivery and Weighted Adverse Outcome Score in each subgroup. Results The overall caesarean rate was 31.2%. Repeat CD contributed 29.5% of all CD, followed by nulliparas with labour induction (15.3%) and non-cephalic presentation (14.3%). The caesarean rates in induced nulliparas with a term singleton cephalic pregnancy and women with previous CD were 31.6% and 82.0%, respectively. CD had no clinically meaningful association with perinatal outcomes in most subgroups. However, in singleton preterm breech presentation and preterm twin gestation with the first twin in non-cephalic presentation, CD was associated with substantially improved maternal and perinatal outcomes. Conclusions Women with repeat CD, term non-cephalic presentation, term twins or other multiple gestation and preterm births may be the potential targets for safely reducing prelabour CD rate, while nulliparas or multiparas with spontaneous or induced labour, women with repeat CD, term non-cephalic presentation, term twins or other multiple gestation and preterm births are potential targets for reducing intrapartum CD rate without compromising maternal and neonatal safety in the USA. On the other hand, CD may still be associated with better perinatal outcomes in women with singleton preterm breech presentation or preterm twins with the first twin in non-cephalic presentation.