Past successes and failures in forecasting criminality are reviewed, and errors which arose in those studies are highlighted. Attention is directed to Greenwood's prediction research as well as a companion RAND study by the Chaikens. The Greenwood study raises the issue of forecasting the rate at which offenders will commit crimes in the future; whereas, most earlier prediction studies examined only whether the offender committed any new crimes. Further, the Greenwood study relied on self-report techniques that may not be reliable. The research produced a 'predictive' instrument that uses many of the same factors that have been used in prediction studies for years: the criminal record, employment history, and drug use. Further, Greenwood's projections of crime reduction were derived from a flawed research design. Also, selective incapacitation, in the form considered in the Greenwood study, does not alleviate the problems of overprediction and undeserved punishment typically associated with predictive sentencing. Although Greenwood asserts that selective incapacitation strategies can alleviate prison overcrowding, analysis does not support his contention. The Chaikens' study uses the same self-report data as did Greenwood, but the Chaikens' study is more cautious in its conclusions. Finally, the Chaikens state that theirs is 'postdiction research' rather than a true prediction study. Included are 139 references.