Selective attention (SA), the process by which information is prioritized for processing according to its relevance to current goals, and working memory (WM), the temporary storage and/or manipulation of information in mind, are considered to be important building blocks in human cognition. Both are essential for coordinating thought and action, and both are foundational for the emergence of other more complex executive functions, like planning and problem solving. The complicated interplay between SA and WM has been investigated across a growing number of experimental studies, with attentional processes influencing various stages of WM, and vice versa. Behavioural evidence suggests that SA can bias processing of information as we anticipate, encode and maintain contents in memory, whilst WM can serve to maintain a template as we search. Neuroimaging studies have observed a highly similar frontoparietal network subserving both processes, indicating anatomical and functional overlap in their corresponding neural mechanisms. Nevertheless, despite substantial evidence for cognitive and neural overlap, almost everything we know about the relationship between SA and WM is derived using group-average performance. In reality, some individuals may rely on shared sub-processes to perform tasks more so than others. In this thesis we extended previous work by understanding this individual variability. The first experimental chapter describes the development of two behavioural paradigms tapping SA and WM. These paradigms are better suited to address this question, relative to previous experimental approaches, because they are matched on task-specific features while being independently scalable in terms of difficulty. The second experimental chapter used functional magnetic resonance imaging (fMRI) in combination with these tasks to identify the neural correlates of individual differences in the strength of SA-WM coupling across participants. The third experimental chapter builds upon the neuroimaging study and addresses whether computational models trained to perform the same set of tasks share any mechanistic properties observed in the human brain, providing a useful framework in which predictions about the relationship between cognitive processes can be readily tested. Lastly, in the final experiment we used cognitive training to test whether altering SA would lead to changes in the related WM system, and whether these gains are modulated by baseline individual differences in the strength of their coupling. Together, along with an opening General Introduction and concluding Discussion, these chapters explore heterogeneity in the relationship between SA and WM from multiple perspectives, integrating advances in human cognition, neuroimaging and computational modelling.