Preprint: We compared the predictive potential of dynamic and static risk factors of recidivism, via machine learning methods. The data contained 746 men that had, and 746 men that had not, reoffended during follow-up periods between 179 and 1332 days. Static predictors included the crime committed, prison history, and age. Dynamic predictors were 50 items from the Finnish Risk and Needs Assessment Form (RITA). Static risk factors strongly predicted both general and violent recidivism. Dynamic predictors performed slightly worse—they added little beyond static risk factors to the prediction of general recidivism and minimally to the prediction of violent recidivism. All the predictive models had good discriminative power with AUC between .70 and .80 and good calibration. Using static predictors, however, produced a wider range of estimated probabilities. Results show that these dynamic risk factors, as assessed, do predict recidivism but, in our opinion, risk assessments should primarily use static predictors.