The predictive value of a child abuse screening instrument on unselected populations is illustrated for varying hypothesized levels of child abuse prevalence to demonstrate outcome of a hypothetical national screening program. At any level of application, the prediction of false positives (nonabusing families labeled as abusing or potentially abusing) and false negatives is seen to suggest low practical utility and an unacceptably large social cost. Varying interpretations of the meaning of this possible misclassification are discussed in the context of several program models. The following hypotheses for misclassification are posed: that subjects are really misclassified due to misdiagnosis based on either misinformation or socially induced bias; that subjects only appear misclassified because of limited ability to distinguish between types of cases; and that subjects are misclassified because diagnostic categories are overlapping (i.e. that child abuse is not a distinct category from accidents). Three short case examples demonstrate the implications of these kinds of misclassification for clinical practice program and policy. (Author/SBH)