Reliable and valid risk assessments are essential for responding adequately to juveniles who have sexually offended (JSO). Given the lack of specific research focusing on the JSO population, the present study aims at confirming and expanding previous findings based on clinical samples dealing with risk assessments of JSO. The predictive power of the Juvenile Sex Offender Assessment Protocol (J-SOAP-II) and the Sexual Offence Severity (SOS) Scale are evaluated retrospectively by analyzing forensic, police, and judicial files in a consecutive sample of 223 adolescents (mean age of 15.7 years, SD = 2.1 years) who had been convicted of a sexual crime in the Canton of Zurich, Switzerland, between 2000 and 2008. Based on local official recidivism data (mean follow-up period = 4.3 years; SD = 2.5 years), univariate and multivariate predictions of sexual and nonsexual recidivism are tested by use of the J-SOAP-II and the SOS Scale in logistic regression and receiver operating characteristics (ROC) analyses. Sexual recidivism (n = 7, 3.1%) is best predicted by a multivariate model, including the J-SOAP-II Adjustment Scale and the SOS Scale (ROC: Area under the curve [AUC] = .818). Nonsexual violent (n = 37, 16.6%) and general recidivism (n = 100, 44.8%) are only moderately predicted by the J-SOAP-II impulsive/antisocial behavior scale (AUC = .677 and .662, respectively). In conclusion, the J-SOAP-II adjustment scale, the J-SOAP-II impulsive/antisocial behavior scale, and the SOS Scale may be helpful for screening purposes in JSO, but additional risk assessment seems necessary. Overall, further research is necessary for a comprehensive understanding of risk factors in JSO.