Risk assessment instruments are increasingly used in mental health jail diversion programs. This study examined the reliability and validity of Short-Term Assessment of Risk and Treatability (START) and Level of Service Inventory-Revised (LSI-R) assessments overall and by client race. Research assistants completed START and LSI-R assessments for 95 diversion clients. Arrests and jail days were collected via official records and self-report 3, 6, 9, 12, and 18 months after baseline. Assessments demonstrated good interrater reliability and convergent validity. START strength total scores and LSI-R risk estimates were the strongest predictors of recidivism. Total scores and risk estimates did not differ as a function of client race, but there were some differences in accuracy of START vulnerability and LSI-R total scores and risk estimates in predicting jail days (but not arrests), over shorter follow-ups. No such differences were found for START strength total scores across any follow-up period or recidivism measure.