Behaviour can be goal-directed, when performing an action to obtain a specific goal, and it can be habitual, whereby a stimulus in the environment can trigger a response regardless of the outcome. Behavioural output can be a mixture of goal-directed and habitual aspects and there may be a competitive balance between these dual influences. Degrading contingencies between actions (A) and their outcomes (O) challenges beliefs about cognitive control and can be used to distinguish between behaviour that is goal-directed or habitual. In this thesis, the possible causal role of specific brain regions in controlling this balance between goal-directed and habitual behaviour, as measured in the contingency degradation paradigm, were determined. As argued in the thesis Introduction, studies of the marmoset help to bridge rodent and human studies, including possible clinical translation, one of the reasons being the homologies that exist for the prefrontal cortex (PFC). A novel touchscreen-based contingency degradation task was developed for the common marmoset monkey, a New World non-human primate. The possible roles of the primate medial PFC (area 32), anterior cingulate cortex (area 24), ventromedial PFC (area 14-25), anterior orbitofrontal cortex (OFC, area 11), medial OFC (area 14) and the caudate nucleus (CN) were then compared following training, using reversible pharmacological inactivation and activation of these structures via implanted cannulae. None of the studies in the literature had examined the causal role of these brain regions in the expression of A-O associations as measured by contingency degradation. Inactivation of either area 24 or the CN significantly impaired the animals’ sensitivity to contingency degradation, as did activation of area 24. These findings suggest that area 24 and the CN may form part of a neural circuit that mediates the expression of A-O contingencies, a hypothesis supported by an anatomical tracing study. By contrast, inactivation of area 11 apparently enhanced sensitivity to instrumental contingency degradation, possibly by blocking competing pavlovian associations. Manipulations of area 14, 25 or area 32 did not affect sensitivity, indicating the neuroanatomical specificity of the contingency degradation deficits. Control experiments ruled out the possible contribution of effects on primary motivation or any non-specific effects of inactivation or activation. The findings are interpreted in the light of literature suggesting that the PFC sub-regions have largely distinctive but overlapping roles in controlling goal-directed behaviour. Additionally, a specific PFC sub-region, area 24, may work together with anterior CN to maintain and utilise learned causal relationships between actions and outcomes. The current thesis’s study on the expression of goal-directed knowledge may be useful for explaining the chronic psychopathology of several psychiatric disorders, including obsessive-compulsive disorder, as well as its possible neural substrates.