Objective: In this longitudinal study, the predictive validity of a psychiatric diagnosis of sexual sadism was compared with three behavioral indicators of sadism: index sexual offense violence, sexual intrusiveness, and phallometrically assessed sexual arousal to depictions of sexual or nonsexual violence. Method: Five hundred and eighty six adult male sexual offenders were assessed between 1982 and 1992, and these offenders were followed for up to 20-years postrelease via official criminal records. Assessment information included the "Diagnostic and Statistical Manual of Mental Disorders" (DSM) diagnosis, offense characteristics, phallometric assessment results, and an actuarial risk measure (the Sex Offender Risk Appraisal Guide). Results: Predictive validity was demonstrated in univariate analyses for the behavioral indicators of sexual sadism (area under the curve [AUCs] from 0.58 to 0.62) but not psychiatric diagnosis (AUC = 0.54). Cox regression analyses revealed that phallometrically assessed sexual arousal to violence was still significantly associated with violent (including sexual) recidivism after actuarially estimated risk to reoffend was controlled. A psychiatric diagnosis of sexual sadism, in contrast, was unrelated to recidivism. Conclusions: The results support the use of more behaviorally operationalized indicators of sexual sadism, especially phallometric assessment of sexual arousal, and suggest the DSM criteria for sexual sadism require further work. (Contains 4 tables.)