Background Hyperglycemia in preterm infants may be associated with severe retinopathy of prematurity (ROP) and other morbidities. However, it is uncertain which concentration of blood glucose is associated with increased risk of tissue damage, with little consensus on the cutoff level to treat hyperglycemia. The objective of our study was to examine the association between hyperglycemia and severe ROP in premature infants. Methods and findings In 2 independent, monocentric cohorts of preterm infants born at <30 weeks’ gestation (Nantes University Hospital, 2006–2016, primary, and Lyon-HFME University Hospital, 2009–2017, validation), we first analyzed the association between severe (stage 3 or higher) ROP and 2 markers of glucose exposure between birth and day 21—maximum value of glycemia (MaxGly 1–21 ) and mean of daily maximum values of glycemia (MeanMaxGly 1–21 )—using logistic regression models. In both the primary ( n = 863 infants, mean gestational age 27.5 ± 1.4 weeks, boys 52.5%; 38 with severe ROP; 54,083 glucose measurements) and the validation cohort ( n = 316 infants, mean gestational age 27.4 ± 1.4 weeks, boys 51.3%), MaxGly 1–21 and MeanMaxGly 1–21 were significantly associated with an increased risk of severe ROP: odds ratio (OR) 1.21 (95% CI 1.14–1.27, p < 0.001) and OR 1.70 (95% CI 1.48–1.94, p < 0.001), respectively, in the primary cohort and OR 1.17 (95% CI 1.05–1.32, p = 0.008) and OR 1.53 (95% CI 1.20–1.95, p < 0.001), respectively, in the validation cohort. These associations remained significant after adjustment for confounders in both cohorts. Second, we identified optimal cutoff values of duration of exposure above each concentration of glycemia between 7 and 13 mmol/l using receiver operating characteristic curve analyses in the primary cohort. Optimal cutoff values for predicting stage 3 or higher ROP were 9, 6, 5, 3, 2, 2, and 1 days above a glycemic threshold of 7, 8, 9, 10, 11, 12, and 13 mmol/l, respectively. Severe exposure was defined as at least 1 exposure above 1 of the optimal cutoffs. Severe ROP was significantly more common in infants with severe exposure in both the primary (10.9% versus 0.6%, p < 0.001) and validation (5.2% versus 0.9%, p = 0.030) cohorts. Finally, we analyzed the association between insulin therapy and severe ROP in a national population-based prospectively recruited cohort (EPIPAGE-2, 2011, n = 1,441, mean gestational age 27.3 ± 1.4, boys 52.5%) using propensity score weighting. Insulin use was significantly associated with severe ROP in overall cohort crude analyses (OR 2.51 [95% CI 1.13–5.58], p = 0.024). Adjustment for inverse propensity score (gestational age, sex, birth weight percentile, multiple birth, spontaneous preterm birth, main pregnancy complications, surfactant therapy, duration of oxygen exposure between birth and day 28, digestive state at day 7, caloric intake at day 7, and highest glycemia during the first week) and duration of oxygen therapy had a large but not significant effect on the association between insulin treatment and severe ROP (OR 0.40 [95% CI 0.13–1.24], p = 0.106). Limitations of this study include its observational nature and, despite the large number of patients included compared to earlier similar studies, the lack of power to analyze the association between insulin use and retinopathy. Conclusions In this study, we observed that exposure to high glucose concentration is an independent risk factor for severe ROP, and we identified cutoff levels that are significantly associated with increased risk. The clinical impact of avoiding exceeding these thresholds to prevent ROP deserves further evaluation.