The neural networks associated with socio-affective (empathy, compassion) and socio-cognitive processes (mentalizing/Theory of Mind) have been well-characterized over the last years. The goal of the present talk is twofold: (1) To explore the separability of these functions during online social understanding on a subjective, behavioral and on a neural level and (2) to investigate the embedding of the related neural substrates in large-scale task-free neural networks. To this end, we acquired resting state as well as behavioral and neuroimaging data (fMRI) during a social video task in a large sample of participants (N = 178). The videos were short autobiographical narrations of emotionally negative and neutral events that allowed for asking Theory of Mind questions about the thoughts of the narrators and factual reasoning questions about the content of the stories, thereby allowing for independent assessment of socio-affective and socio-cognitive processing. Linking the phenomenological with the neural level, participants reported increased negative affect after emotional stories, which covaried with activity strength in the meta-analytically defined “empathy network”, but not with activity in the “Theory of Mind network”. Vice versa, performance in answering the Theory of Mind questions correlated with “Theory of Mind network”, but not “empathy network” activity. Interestingly, neither behavioral markers of social affect and mentalizing (i.e. emotional valence ratings and Theory of Mind performance) nor activity in the two respective neural networks correlated with each other. Furthermore, resting state functional connectivity to task activation based seed regions for empathy and Theory of Mind yielded distinct networks that strongly overlapped with the respective task activations and correspond to the well-described default mode network (Theory of Mind seeds) and the salience or central executive network (empathy). The data strongly argue for dissociable and independent socio-affective and -cognitive functions that are embedded in large-scale task-unrelated neural circuits.