Men who sexually offend against children often deny their offences outright, or attempt to present themselves as more socially acceptable by minimisation and rationalisation of the offence. Professionals need accurate information about the beliefs and attitudes that underpin this type of offending. Equally, an accurate, easily administered and cost-effective means to indirecdy assess the levels of risk these men would present if released back into society is vital. Currendy, statistical combination of past-offence data and demographic information produces actuarial risk prediction. This has been criticised as insensitive to changes in levels of risk. Also, existing dynamic risk assessments tend to rely on self-report and interview, both of which may be open to the self-presentational efforts of men in denial. In response to these shortcomings, the research presented in this thesis represents an attempt to develop the Implicit Association Test (IA T; Greenwald, McGhee & Schwartz, 1998) for application specifically to this offender population, and to test its diagnostic accuracy in differentiating paedophilic sex offenders from non-sex offenders. The original IAT format had been shown to identify offenders against minors from other offenders (Gray, Brown, Smith & Snowden, 2005). However, feedback from participants suggested the task might be subject to the effects of low IQ and poor literacy in this population. Hence, this body of work modified the standard task to create an offender-specific lAT. To achieve this, experiment 1 first identified equivalent lAT effects for words and for picture stimuli. Experiment 2 confirmed the "IA T effect" in a shortened 2-stage task, and confirmed that order effects were negligible, but that the number of exemplars per IA T category had some impact on IA T effect. A streamlined child-sex association IAT (CSA-IA1) was created based on these findings. Experiment 3 applied the CSA-IAT to community-based samples of men convicted of sexual offences against children and non-sex offenders, and showed it differentiated the groups with very good diagnostic accuracy. Questions remain in the literature regarding the "fakeability" of the IA T. Experiments 4 and 5 investigated faking of two variants of the novel task (based on nationality and sexuality), and identified a pattern of faking which supported, but extended, previous findings. Experiment 6 then tested the new IA T in the prison population. Inmates convicted of paedophilic sexual offences were compared to men convicted of hebephilic sexual offences and men convicted of non-sexual offences. The task discriminated among the sexual offenders, and differentiated paedophilic offenders from controls, with excellent diagnostic accuracy. It was unaffected by IQ, age and cognitive ability. Finally experiment 7 applied the new test to paedophilic offenders who deny their offences. The CSA-IAT showed very good diagnostic accuracy at discriminating paedophilic deniers from controls, with ROC values for the denier/control discrimination equivalent to those for the admitter/ control discrimination. In summary, this thesis identifies (i) the potential of the IA T as a tool to assist professionals by increasing our understanding of the cognitions underpinning sexually-offending behaviour, and (ii) the potential of the task to be developed as a means to indirectly measure levels of dynamic risk of sexually offending against children presented by convicted abusers.