Background In Sweden, the number of people serving life sentences has steadily increased. To date, few studies have examined the recidivism rate or the predictive validity of different risk assessment instruments in this group. Aims Our aim was to test the predictive validity among inmates serving life sentences of two different instruments used for assessing risk—the Historical, Clinical and Risk Management‐20 (HCR‐20), most widely used in clinical populations, and the Psychopathy Checklist‐Revised (PCL‐R), commonly applied in both penal and clinical settings. Method Ninety‐eight life‐sentenced prisoners were included, 26 of whom were released during the study period. Data on risk assessments and dates for release were collected from administrative records, while recidivism data were retrieved from a national database on criminal convictions. Results Sex offenders obtained the highest scores and inmates charged with domestic violent offences obtained the lowest scores on both instruments. The released prisoners were followed for a mean period of 33 months. During this time five prisoners (19%) reoffended, four of them violently, with an average time to recidivism of 10 months. Only PCL‐R Facet 4, which reflects antisocial features, was significantly associated with recidivism. Conclusions This small, but population‐based, study demonstrates that antisocial behaviour shows incremental predictive validity for reoffending among life‐sentenced offenders, but other measures have little to add for this specific task. The fact that those life sentenced prisoners who reoffended did so so soon after release should prompt allocation of earlier interventions towards preventing this. Copyright © 2015 John Wiley & Sons, Ltd.