A fundamental question of human existence is how much control we have on our behaviour. This dissertation aims to add to our understanding of cognitive control by characterising how a particular failure of performance, Goal Neglect (GN), is affected by different forms of complexity manipulations. In Chapter 2, I develop a new task to test GN and unlike previous studies, I manipulate complexity qualitatively by altering the instructional cues - the cues instructing the participant to shift to a different rule set. GN was sensitive to this kind of complexity manipulation and this is linked to a failure in recognizing the significance of the instructional cues. In Chapter 3, I propose a new entropy-like measure to quantify the temporal clustering of GN and use this to test the differential temporal patterns that are predicted by two theoretical models of GN. The results suggest that both models are likely to be operant, but with their relative dominance being different across time: GN early on in the task appears to be mostly driven by failures which are “task model” like, whilst GN which manifests later on is better aligned with the “monitoring” account. Chapter 2 also revealed that GN can be sensitive to manipulations of complexity during task performance, which motivated the question of whether previously published studies suggesting the contrary, were perhaps due to insufficient complexity. Hence, in Chapter 4, using the new GN task, I investigate this further. Overall, the results were mixed and indicated that complexity does not appear to affect GN unless the complexity manipulation is more closely associated to the critical event. Throughout this dissertation, I refer to models and empirical evidence from the Prospective Memory (PM) literature given the apparent similarity between PM and GN experimental paradigms. In Chapter 5, I take this further and investigate how PM failures and GN are different, if at all, with the broader aim to integrate what are otherwise isolated domains. I found a mixture of null findings which suggest that it is not entirely clear if GN and PMf reflect different capacities. Nonetheless, while investigating the differences between GN and PMf, a much more interesting question emerged with respect to what structural features of a task predict different signatures of GN-like and PMf-like errors. The key finding to this theory-neutral approach was a general rule about task structure: a combination of extended practice and low frequency of critical events predict both a larger amount of errors and with more of these occurring late in the task. Overall, this research has shed further light on task conditions that may result in different error signatures and that may reflect different cognitive resources.