The Criminal Sentiments Scale (CSS) purports to measure attitudes and values pertaining to and in support of criminal conduct (Andrews & Bonta, 2003). A sample of 72 sex offenders treated in a high intensity sex offender program completed the CSS pre- and post-treatment and was followed up for approximately 3 years. The CSS demonstrated high internal consistency. Correlational and ROC analyses indicated that the CSS is a robust predictor of non-sexual violent and non-violent recidivism. The CSS also correlated significantly with the General Statistical Information on Recidivism Scale which assesses non-sexual recidivism risk. Similar analyses showed that the CSS did not predict sexual recidivism nor did it correlate with a tool that assesses sexual recidivism risk—the Static 99. The CSS scores changed significantly in the predicted direction following treatment. The present results suggest that the CSS is an internally consistent tool for the assessment of criminal sentiments among sex offenders and that it predicts non-sexual but not sexual recidivism.