Social neuroscience has identified some of the key brain structures involved in social perception and cognition. Much less is known about the underlying neural dynamics and inter-individual variability during real time social interaction. The Virtual Partner Interaction (VPI) paradigm is a “human dynamic clamp” consisting of a human subject interacting reciprocally and in real-time with a computational model of him/herself. The virtual partner (VP) is based on am empirically validated model of human coordination behavior and is construed as having a degree of autonomy. Such a surrogate system for human social coordination allows parametric manipulation of the intrinsic dynamics of the virtual partner and its coupling to the human in real-time. We view social interaction as more than (and different from) the sum of individual behaviors and intentions. We assess subjects’ ability to recognize the cooperative or competitive intentions of their partner and by doing so uncover underlying behavioral factors and their neural dynamics. Subjects were instructed to coordinate continuous finger movements in-phase or anti-phase with the VP, while the latter maintained either a cooperative (same as human) or competitive (opposite to human) behavior, or switched between the two. The behaviors of both actual and virtual partners were recorded continuously, and subjects verbally reported VP’s intentions. We demonstrated that strong coupling, competitiveness and changes of intention of the VP improved the human’s attribution of intention. During the task, high-density EEG (121 electrodes) was recorded with the goal of determining the neural dynamics of the process of intention attribution. Coordination with the virtual partner elicited brain rhythms within the 10Hz range over the right parietal region. Statistical group analysis confirmed this neural link with intention attribution. Behavioral and neural dynamics revealed different subgroups with highly reproducible task-related neural signatures. These results demonstrate how standard-averaging methods can erase functionally relevant dynamics at the intra-individual level. A full account of intention attribution thus needs to consider the different reproducible neural signatures expressed within individuals. This is specifically of matter of importance for the diagnosis of mental disorders such as autism and schizophrenia, which combine impairment of intention attribution and high inter-individual variability.