In response to the risk of serious further offences, an evidence-based approach is needed in risk management. A recent joint prison–probation inspection of the management of life sentence prisoners in six U.K. prisons found that the quality of assessment and plans to manage risk of harm to others was insufficient, with too much focus on the offender's verbal account. The present paper discusses observations of regular prisoner behaviour as the basis for predictions, and summarizes results of an evaluation of this methodology based on a sample of high-risk category prisoners released into the community. Prison behaviour has not traditionally been seen as a valid risk marker for violent recidivism, which may be because typically only conspicuous high-level behaviours are considered by risk management panels. Our research suggests that we are neglecting a valuable source of information on risk by failing to observe on-going and consistent pre-release behaviour.