To compensate for insufficient resources, many child protection agenci
es use screening instruments to prioritize reports of abuse for invest
igations. Like most risk assessment instruments, however, the variable
s included in the screens are selected on the basis of clinical observ
ations. Because an empirically derived screen may lead to more-accurat
e predictions, this article identifies predictors of injury severity c
aused by physical abuse from information included in the abuse reports
. Subjects included one physically abused child from 789 families. The
criterion variable had four levels: no, mild, moderate, and severe in
jury. Predictors included child, report, perpetrator, and maternal cha
racteristics. Results of an ordinal probit regression analysis identif
ied that a model with four predictors (perpetrator identity, reporter
identity, severity of allegations, and season report was mode) and two
interaction terms (child's age by mother's age and child's age by chi
ld's gender) successfully predicted whether a third was injured.