The social class gradient in childhood injury mortality is steep and increasing, so there is emphasis on targeting injury prevention on the basis of socioeconomic deprivation, to reduce inequalities in health. This paper examines the relationship between medically attended unintentional injury, sociodemographic characteristics and previous injury. This was a cohort study using the control group from a cluster randomised controlled trial of injury prevention in primary care. The cohort comprised children aged 3-12, months registered with participating practices, whose parents completed the baseline questionnaire (n = 771). 94% were followed for 25 months. Medically attended unintentional injury was ascertained from the primary and secondary care records. Logistic regression analysis examined the relationship between sociodemographic factors, previous injury and the occurrence of future medically attended injuries. Poisson regression examined the relationship between sociodemographic factors, previous injury and the number of future medically attended injuries. The response rate to the questionnaire was 75%. Residence in a deprived ward, lack of access to a car and male sex were associated with at least one medically attended injury. Residence in a deprived ward and young maternal age were associated with hospital admission. Residence in a deprived ward, male sex and non-ownership of a car were independently associated with number of unintentional injuries. Specificity exceeded sensitivity for all factors for medically attended injury and hospital admission. The positive predictive value was low for all factors, especially for hospital admissions. In conclusion, residence in a deprived ward was independently associated with any medically attended injury, with hospital admission and with number of injuries received. However, more than half of those children residing in a deprived ward did not have a medically attended injury and more than 90% did not have a hospital admission. 60% of children who had a medically attended injury and 40% who had a hospital admission do not live in a deprived ward. A combination of a population approach and targeted interventions will achieve the greatest health gain, and is unlikely to widen inequalities in health.