Important as it is both to risk of re-offending and to cognitive behavioural treatment, violent cognition is seldom measured in rehabilitation programmes, and even more rarely linked to measures of violence risk. Most often, researchers measure violent cognition by having offenders complete transparent self-report questionnaires. This approach may be flawed both by socially desirable responding and by theoretical speculation that stronger links exist between automatic rather than explicit, consciously deliberated cognition and violent behaviour. We measured violent cognition in several ways; collecting data with two self-report scales, along with two Implicit Association Tests (IATs) from men commencing and completing an intensive cognitive-behavioural rehabilitation programme for high-risk violent prisoners. We addressed the questions of whether these two forms of assessment - explicit and implicit - are related, and which is most strongly linked to estimates of violence, based on the Violence Risk Scale. Explicit and implicit tests were not related to each other, although both self-report scales, and one of the IATs elicited significantly more pro-social responses following treatment. Further, the Aggression Questionnaire (AQ) scores were significantly correlated with dynamic risk both pre- and post-programme, while post-programme, scores on one of the two IATs was significantly correlated with dynamic and static risk, as measured pre- and post-programme. These findings suggest that implicit and explicit measures may be assessing different aspects of cognition, and only some are related to violence risk.