Epidemiological research has shown that the majority of children who need professional mental health care do not receive it. Factors other than the level of the child's problem behaviour are involved in the help-seeking process, and are co-determinants of whether or not the child receives help. Longitudinal general population studies have shown that child psychopathology can persist and may lead to later serious adjustment problems, even in adult life. Consequently, it seems important that child psychologists and child psychiatrists should not wait passively until the patient steps into the office, but should actively trace high-risk children and apply the most effective intervention programmes. But how can children most in need of professional interventions be selected accurately? The central question addressed in the Annotation by Bennett et al. in this issue concerns externalising problems. The authors give an account of the most relevant epidemiological measures needed for the assessment of the predictive accuracy of externalising behaviours. They reviewed studies and tested the accuracy with which measures of externalising behaviours in children from the general population predict later antisocial behaviours. Selecting only those studies that pertained to kindergarten and grade one children, they came to the sobering conclusion that the predictive accuracy of externalising symptoms at that age has been overestimated and is, at best, modest. However, as they point out, targeted intervention programmes, in which high-risk children are selected, may still offer some advantages, because even with far from accurate predictive measures the number of children who receive unnecessary interventions can still be substantially reduced. The authors rightly note the risk of labelling due to false positive errors. They also claim that, when the same method is used to detect the presence of externalising behaviours in both females and males, a higher rate of false positive errors in females will occur, with a consequently greater risk for labelling for females. However, they seem to have overlooked that it is possible to select different morbidity or high-risk criteria for each sex separately; by using norm-based measures of problem behaviour, we can use the distributions of problem behaviours in the norm population for each sex separately to determine which cut-off separates high- from low-risk boys and girls. Even if we have improved our means to select accurately those children that run a high risk for later problems, we still have to determine which preventive intervention strategies work best, and who is going to treat all the high-risk children thus selected.