Compulsivity is a transdiagnostic construct, central to mental illnesses including obsessive-compulsive disorder and addictions. The symptoms that define compulsivity are highly overlapping across and heterogeneous within behaviours. It is unknown whether coherent subtypes exist which span across multiple behaviours and are expressed functionally (in underlying cognition and neurobiological function), and clinically (in psychological symptom severity). 45 participants (18–45 years) exhibiting mild-moderate levels of compulsive behaviour in the areas of alcohol use, eating, cleaning, checking or symmetry were recruited from the general community. They were assessed across multiple domains of function including symptom severity, psychological motivation and coping, cognitive performance, neurohormonal activation and brain functional connectivity. Using data-driven machine learning, three distinct subtypes emerged, characterised by good coping (n = 14), positive anticipation (n = 18) and poor coping (n = 13). Subtypes were expressed through differences on self-reported compulsive behaviour severity, experiential avoidance and stress; neurocognitive valence biases in learning; and the cortisol awakening response. Subtypes were compared on amygdala-seed resting-state connectivity, which revealed subgroup differences in systems involved in the bottom-up and top-down generation and regulation of emotion. Findings offer a framework for identifying subtypes that are coherent and meaningful, and cut across traditional boundaries of classical diagnoses.