Objectives: To develop an instrument to determine the probability of child sexual abuse and to estimate the questionnaire s discriminant validity.

Methods: Case-control study of 201 children seen at pediatric clinics and referral centers for the victims of sexual violence, between March and November 2004. Cases comprised children who had either been reported or suspected of being sexually abused and controls comprised children with no such suspicion. We applied a questionnaire to a parent or guardian of each child that consisted of 18 items, each with five Likert scale responses, dealing with the behavior and physical and emotional symptoms exhibited by the children. We excluded nine children for lack of sphincter control. One question was discarded since very few people replied to it. We evaluated the discriminant validity and internal consistency of the items, calculating correlation coefficients (Pearson, Spearman and Goodman-Kruskal), Cronbach sa coefficient and area under the ROC curve. We calculated likelihood ratios and positive predictive values for the five items on the questionnaire that performed best. Results: The questionnaire comprises the five items that best discriminate sexually abused children from nonabused children in two contexts. The score resulting from the sum of responses weighted at 0 to 4 points (overall amplitude of 0 to 20) indicates the post-test probability of sexual abuse by means of the application of Bayes theorem (likelihood ratios).

Conclusions: The instrument is easy to apply and helps in the identification of sexual abuse victims. A cutoff point was defined to indicate the probability of sexual abuse, which can be very useful to guide management of children.