Attempts to link brain data, or behavioural assays thereof, to discrete psychiatric diagnostic categories have been met with limited success in small-scale, case-control studies. An alternative approach is to set aside diagnostic categories and instead use neurobiology and cognition to define more homogenous psychiatric phenotypes. This talk will start by describing traditional case-control designs that were used to identify a putative neurocognitive model of obsessive-compulsive disorder (OCD). Specifically, data suggesting that failures in ‘goal-directed control’ lead to overreliance on the brain’s habit system in OCD. Next, some fundamental limitations of these findings will be discussed in the context of diagnostic specificity, heterogeneity and comorbidity. I will argue that this problem is ubiquitous in psychiatry research and that large-datasets are essential to address it. Focusing on one promising methodology, Internet-based crowdsourcing, I will present data from ~3000 subjects, providing evidence that failures in goal-directed control are related to a trans-diagnostic psychiatric dimension. Finally, I will show data demonstrating that this is cognitively dissociable from other trans-diagnostic dimension, such as anxious-depression.