Objective: To identify achieved oxygen saturations (SpO2) associated with increased risk of severe retinopathy of prematurity (ROP). Design: This is a secondary analysis of the Surfactant Positive Airway Pressure and Pulse Oximetry Trial (SUPPORT)randomised controlled trial. SpO2 was recorded up to 36 weeks' postmenstrual age. Saturations through 9 postnatal weeks were explored graphically, and logistic regression models were created to predict severe ROP. Setting: 20 centres of the Eunice Kennedy Shriver National Institute of Child Health and Human Development Neonatal Research Network. Patients: 984 surviving infants of 24-27 weeks' gestational age born in 2005-2009. Interventions: SUPPORT targeted SpO2 to a lower (85%-89%) or higher (91%-95%) range through 36 weeks' postmenstrual age or off respiratory support. Main outcome measures: Severe ROP defined as threshold ROP, ophthalmological surgery or bevacizumab treatment. Results: There were statistically significant interactions between duration of oxygen supplementation and percentage of time in certain achieved saturation ranges. Specifically, for infants who spent at least 2 weeks on oxygen in postnatal weeks 1-5, a higher percentage of time at 91%-96% SpO2 was associated with increased odds of severe ROP. For infants who spent at least 3 weeks on oxygen in postnatal weeks 6-9, a higher percentage of time at 97%-100% SpO2 was associated with increased odds of severe ROP. Other significant risk factors were lower gestational age and birth weight, non-Hispanic white versus black race, prospectively defined severe illness, late-onset sepsis or meningitis, and clinical centre. Conclusions: Among extremely preterm survivors to discharge, the association between SpO2 and severe ROP depended on the timing and duration of oxygen supplementation.