Objective To assess the value of neonatal EEG for predicting non-optimal neurodevelopmental outcomes in very preterm infants, using a multimodal strategy of evaluation comprising brain imaging and clinical assessment. Design and setting Between 2003 and 2009, we performed an observational, population-based study. Out of 2040 eligible preterm infants born before 32 weeks, 1954 were enrolled in the French regional Loire Infant Follow-Up Team (LIFT) cohort. 1744 (89%) of these completed the follow-up. Neonatal EEGs were recorded prospectively as two EEGs during the first 2 weeks of life and then one every 2 weeks up to 33 weeks. Main outcome measures The neurodevelopmental outcome was assessed by physical examination, the Brunet–Lézine Test and/or the Age and Stages Questionnaire at 2 years of corrected age. Results Of the 1744 infants assessed at 2 years, 422 had a non-optimal outcome. A total of 4804 EEGs were performed, and 1345 infants had at least one EEG. EEG abnormalities were predictive of non-optimal outcomes after controlling for confounding factors such as severe intracranial lesions detected by brain imaging. Transient moderate and severe abnormalities were independent predictors of non-optimal outcomes with an OR and 95% CI of 1.49 (1.08 to 2.04) and 2.38 (1.49 to 3.81), respectively. In the validation group, the predictive risk stratification tree identified severe abnormalities as a factor contributing to the prognosis of two subgroups: infants with severe cranial lesions and infants with a normal examination at discharge and without severe cranial lesions.