Such a strategy proposes that sentence lengths be determined on the basis of how much crime the offender is predicted to commit if not in prison. Predicted high-rate offenders would serve more time than now required, predicted low-rate offenders less. The data do not justify a selective-incapacitation sentencing strategy, however. Predictive validity was modest; crime specialization was relatively rare; specializations did not generally increase with greater numbers of transitions; and the next arrest, regardless of offense category, was likely to be for a nuisance offense. The authors propose a policy of selective deinstitutionalization, which would reduce prison populations without increasing public endangerment. Offenders predicted to be at low-risk of committing a future serious crime would not be imprisoned, and offenders judged to be at high risk of subsequent dangerous crimes would be imprisoned for no longer than offense seriousness would dictate. 40 tables and 52 figures