Aim This study investigated prediction of separate cognitive abilities at the age of 5years by cognitive development at the ages of both 2 and 3years, and the agreement between these measurements, in very preterm children. Methods Preterm children (n=102; 44 males; 58 females) with a gestational age less than 30weeks and/or birthweight less than 1000g were assessed at the ages of 2 and 3years using the second edition of the Bayley Scales of Infant Development, the Child Behaviour Checklist, and a neurological examination, and at the age of 5years using the third edition of the Wechsler Preschool and Primary Scale of Intelligence. Results Cognitive development at ages 2 and 3years explained 44% and 57% respectively of full-scale intelligence at the age of 5years. Adding psychomotor, neurological, and behavioural outcomes to the regression model could not or only marginally improve the prediction; adding perinatal and sociodemographic characteristics to the regression model increased the explained variance to 57% and 64% respectively. These percentages were comparable for verbal intelligence. Processing speed quotient and especially performance intelligence were predicted less accurately. Interpretation Not all aspects of intelligence are predicted sufficiently by the Mental Development Index at ages 2 and 3years. Follow-up of very preterm children until at least the age of 5years is needed to distinguish between different aspects of cognitive development. © The Authors. Developmental Medicine & Child Neurology © 2011 Mac Keith Press.