This study investigates the capacity of pre/perinatal factors to predict attention-deficit/hyperactivity disorder (ADHD) symptoms in childhood. It also explores whether predictive accuracy of a pre/perinatal model varies for different groups in the population. We used the ABCD (Adolescent Brain Cognitive Development) cohort from the United States ( = 9975). Pre/perinatal information and the Child Behavior Checklist were reported by the parent when the child was aged 9-10. Forty variables which are generally known by birth were input as potential predictors including maternal substance-use, obstetric complications and child demographics. Elastic net regression with 5-fold validation was performed, and subsequently stratified by sex, race/ethnicity, household income and parental psychopathology. Seventeen pre/perinatal variables were identified as robust predictors of ADHD symptoms in this cohort. The model explained just 8.13% of the variance in ADHD symptoms on average (95% CI = 5.6%-11.5%). Predictive accuracy of the model varied significantly by subgroup, particularly across income groups, and several pre/perinatal factors appeared to be sex-specific. Results suggest we may be able to predict childhood ADHD symptoms with modest accuracy from birth. This study needs to be replicated using prospectively measured pre/perinatal data.