The predictive validity of four risk assessment instruments: the RRASOR, SVR-20, RM2000-V and the ARMIDILO-Stable and -Acute dynamic client subscales were assessed on a sample of 88 offenders: 44 mainstream and 44 sexual offenders with special needs, who had been matched on risk items within the RRASOR tool. Instruments were coded retrospectively from file information. Sexual reconviction data was used, in conjunction with sexual recidivism data based on unofficial data sources, over a mean follow-up period of 8.8 years. The results of this study found that the ARMIDILO instrument was the best predictor for sexual reconviction among offenders with special needs (ARMIDILO-Stable, AUC=0.60; ARMIDILO-Acute, AUC=0.73), while the predictive validities of the RRASOR (AUC=0.53) and the RM2000-V (AUC=0.50) were little better than chance. In contrast, the SVR-20 yielded a higher score (AUC=0.73) for the non-ID sample, than for the intellectually disabled sample (AUC=0.45). Within the special needs group, the ARMIDILO-Acute, SVR-20 Psychosocial Affect, and Overall scales were better predictors of sexual recidivism for the intellectually disabled subgroup (AUCs ranging from 0.75 to 0.88). These results are discussed in the context of current practice.