THE STUDY WAS PROMPTED BY A REQUEST FROM THE NEW ZEALAND DEPARTMENT OF JUSTICE, WHICH FELT THAT IF LIKELY RECIDIVISTS COULD BE IDENTIFIED EARLY IN THEIR SENTENCES, THEY COULD RECEIVE INTENSIVE REHABILITATIVE ATTENTION. THE STUDY INVOLVED THE CONSTRUCTION AND VALIDATION OF AN EXPERIENCE TABLE SUMMARIZING THE POSTRELEASE HISTORIES OF A SAMPLE OF 347 INMATES. THE INMATES WERE FOLLOWED FOR 3 YEARS AFTER THEIR RELEASE AND CODED AS SUBSEQUENTLY RECONVICTED OR NOT RECONVICTED. THREE METHODS OF PREDICTION ARE TESTED: (1) A POINT-SCORE SYSTEM, WHICH ASSIGNS INMATES POINTS FOR EACH CHARACTERISTIC THEY POSSESS THAT CORRELATES POSITIVELY WITH RECONVICTION IN THE ORIGINAL SAMPLE; (2) DISCRIMINANT FUNCTION ANALYSIS, A REFINEMENT OF THE POINT-SCORE SYSTEM; AND (3) ANALYSIS BY AUTOMATIC INTERACTION DETECTOR. VARIABLES USED IN THE DISCRIMINANT FUNCTION ANALYSIS INCLUDE TYPE OF OFFENSE, DRINKING AT TIME OF OFFENSE, AGE AT CONVICTION, PREVIOUS COURT APPEARANCES, NATIONALITY, MARITAL STATUS, AND EMPLOYMENT. METHODOLOGICAL PROBLEMS ARE NOTED. IT IS CONCLUDED THAT, REGARDLESS OF THE METHOD OF PREDICTION USED, UNKNOWN OR UNMEASURABLE FACTORS (E.G., THE DEGREE OF INTEREST TAKEN IN AN OFFENDER OR THE OFFENDER'S OWN DECISION TO GIVE UP CRIME) INFLUENCE OUTCOME AND INTERFERE WITH RELATIONSHIPS THAT APPEAR TO EXIST BETWEEN CERTAIN VARIABLES AND RECONVICTION FOR THE POPULATION AS A WHOLE.