Predictive risk modelling is being developed in New Zealand to identify children at high risk of maltreatment and provide preventative services. However, the accuracy of predictions from models tested so far has been too low for use as a sole mechanism. The main predictors identified have been a caregiver's history of welfare benefits, single parenthood status, and their own history of maltreatment or other children's maltreatment reports. However, these predictors are strongly associated with poverty and may serve to identify many poor families rather than justify child welfare intervention. Improving data quality and developing new datasets specifically for modelling may help increase predictive ability in the future.